Khaled Alshehri,a Marwa Mahmoud AlFattani,b Laila Bashmal,c and Ghalia Alshmmarid

December 2025 | Doi: 10.30573/KS--2025-DP69

AI and Energy: The Future of

Data Centers in Saudi Arabia

aFellow, King Abdullah Petroleum Studies and Research Center (KAPSARC); bFellow, International Center for AI Research & Ethics (ICAIRE);

cResearcher, King Saud University (KSU); dFellow, ICAIRE

Discussion Paper

2AI and Energy: The Future of Data Centers in Saudi Arabia

About KAPSARC

About ICAIRE

KAPSARC is an advisory think tank within global energy

economics and sustainability providing advisory services to

entities and authorities in the Saudi energy sector to advance

Saudi Arabia’s energy sector and inform global policies through

evidence-based advice and applied research.

International Center for Artiﬁcial Intelligence Research and

Ethics (ICAIRE) — Under the Auspices of UNESCO

ICAIRE serves as a global “lighthouse” for the advancement

of ethical AI. The Center is dedicated to ensuring that AI

technologies are developed and deployed in harmony with

human rights and moral integrity. This is achieved through

international cooperation to coordinate AI research and

development, raise awareness about AI ethics, foster specialized

skills, and provide expert advisory support on AI policies.

This publication is also available in Arabic.

Intelligence Research and Ethics (“ICAIRE”). This Document (and any information, data or materials contained therein) (the

“Document”) shall not be used without the proper attribution to KAPSARC and ICAIRE. The Document shall not be reproduced, in

whole or in part, without the written permission of KAPSARC and ICAIRE. KAPSARC and ICAIRE make no warranty, representation or

undertaking whether expressed or implied, nor do they assume any legal liability, whether direct or indirect, or responsibility for the

accuracy, completeness, or usefulness of any information that is contained in the Document. Nothing in the Document constitutes

or shall be implied to constitute advice, recommendation or opinion. The views and opinions expressed in this publication are those

of the authors and do not necessarily reﬂect the ofﬁcial views or position of KAPSARC or ICAIRE.

Legal Notice

KEY

POINTS

Under a high-growth scenario, Saudi Arabia could exceed

4 GW of data center capacity by 2030, emerging as a

regional AI compute hub.

AI data centers could consume up to 11% of national

electricity by 2030 under high-growth scenarios.

Utilization rates and hardware efficiency have a greater

impact on AI data center project costs than electricity

tariffs.

The Kingdom’s digital and energy infrastructure is

central to AI readiness, data sovereignty, and economic

diversification.

Strategic alignment of AI and energy policy positions

Saudi Arabia as a competitive, climate-aware AI host

nation.

Executive Summary

The rise of artiﬁcial intelligence (AI) is rapidly transforming the global economy,

making AI-ready data centers essential drivers of this change. Unlike traditional

data centers for general information technology (IT), AI-focused facilities use advanced

chips, dense servers, and liquid cooling to support high-performance computing. This

shift accelerates digital innovation but increases pressure on energy systems. Globally in

2024, total data center capacity exceeded 111,900 MW, with the United States and China

accounting for more than 60%. Capacity is projected to double to 224,000 MW by 2030,

with electricity use rising from 854 TWh in 2024 to nearly 1,900 TWh by 2030. AI workloads

already use 5%-15% of that power and could reach 35%-50% by the end of the decade,

highlighting the growing link between the energy and digital sectors.

Saudi Arabia’s data center industry is expanding faster

than most regions, driven by coordinated investment in AI

infrastructure.

By 2024, the Kingdom had 58 operational facilities with a total

IT capacity of 290.5 MW, primarily in Riyadh and Dammam,

which account for nearly 80% of national data center capacity.

New hubs, including NEOM, are emerging for large-scale AI

projects. The sector’s growth is supported by progressive

digital policies like the Cloud First Policy and the Data Center

Services Regulations, alongside expanding grid infrastructure

and low-cost energy. This base makes Saudi Arabia the largest

digital infrastructure market in the Middle East and one of the

few countries globally developing AI-optimized campuses at a

multi-gigawatt scale.

Electricity demand from Saudi data centers is expected

to rise substantially until 2030, but actual outcomes are

uncertain.

National data centers used around 2.8 TWh in 2024, or 0.85%

of total electricity. By 2030, this could increase to between

10.2 TWh and 42.2 TWh, representing 2.8%-11.6% of projected

national electricity demand. The increases in demand reﬂect

a total installed capacity of roughly 2,000 MW under a

moderate-growth scenario and up to 4,100 MW under a high-

growth case. However, many global analysts view the pace of

AI infrastructure expansion as highly uncertain – potentially

resembling a “digital infrastructure bubble” in which some

announced projects do not materialize or are delayed. The

low-growth scenario (around 1,050 MW) remains a credible and

conservative planning baseline.

The energy and environmental implications of this growth

are signiﬁcant and deserve careful management.

Under a fossil-fuel-dominated power mix, emissions from

data centers could rise from 1.6 Mt CO2 in 2024 to 6-24 Mt

CO2 by 2030. While this is a small share of the Kingdom’s total

emissions, currently estimated at around 590 Mt CO2 per year,

the growth trajectory highlights the need to include new digital

loads within the broader decarbonization strategy. Achieving

the national target of 50% renewable generation could reduce

data center emissions by about 68%. Efﬁciency improvements

could further mitigate the impact, reducing their electricity

consumption by 13% and saving up to 5 TWh annually in the

high-growth scenario. From an energy system perspective,

these facilities may act as new baseload consumers, and their

rapid build-out requires close coordination with utilities to

ensure grid reliability and adequate capacity, and to prevent

localized bottlenecks.

Saudi Arabia has notable cost advantages which could be

further sustained through efﬁcient operations and careful

tariff management.

The study’s cost analysis shows that data center project

costs in Saudi Arabia are most sensitive to utilization and

hardware efﬁciency, and moderately sensitive to electricity

tariffs and power usage effectiveness (PUE). With low

AI and Energy: The Future of Data Centers in Saudi Arabia

competitive tariffs and expanding grid infrastructure, Saudi

Arabia remains globally competitive even with relatively

high cooling requirements. This competitiveness depends on

stable tariffs, early high-utilization rates, and energy-efﬁcient

hardware. If these fundamentals are sustained, the Kingdom

could become a regional hub for AI computing, serving both

domestic and cross-border digital demand. Policymakers can

enhance competitiveness by introducing efﬁciency standards

and encouraging high-performance equipment. The analysis

also shows that most cost gains occur as data centers move

from partial to steady utilization, with diminishing returns at

high load factors, a key consideration for utilities and investors

optimizing grid integration and cost efﬁciency.

Despite its advantages, Saudi Arabia faces similar risks to

other rapidly expanding markets.

The global AI data center boom has raised regulatory,

environmental, and ﬁnancial concerns. Projects need large

tracts of land, highly skilled technical labor, and secure access to

reliable electricity and water for cooling. Rising hardware costs,

supply chain constraints, and geopolitical risks add further

uncertainty. Financially, the surge in AI-related investment

resembles a speculative “gold rush,” where capital chases

uncertain demand. For Saudi Arabia, this highlights the need to

sequence projects, aligning expansion with realistic utilization

forecasts, and integrating new loads into national energy

planning to avoid overcapacity or stranded assets.

Sustainable strategies can mitigate many of these risks and

strengthen competitiveness. Technologies such as modular

design, AI-optimized chips, and advanced liquid or water-free

cooling systems can sharply improve efﬁciency. Workload

scheduling and AI-based energy management can reduce

operational loads, while renewable energy integration through

power purchase agreements and 24/7 carbon-free energy

matching is emerging. Such integration also strengthens Saudi

Arabia’s positioning as a clean-energy AI hub, aligning digital

infrastructure growth with the Kingdom’s broader energy

transition. Leading global companies demonstrate this shift:

Microsoft’s waterless cooling, Google’s geothermal supply,

Amazon Web Services’s (AWS’s) 100% renewable procurement,

and Meta’s heat recovery systems illustrate best practice in the

industry.

The rationale for investing in AI-ready data centers goes

beyond short-term returns and needs careful prioritization

and coordination. These facilities create digital spillovers

that strengthen the Kingdom’s innovation ecosystem, data

sovereignty, and economic diversiﬁcation. Sustained value

depends on investing in AI-ready zones with reliable power,

renewable integration, and strong utilization. Coordinated

planning between government entities, private developers, and

global technology partners is essential to maximize beneﬁts

while ensuring energy security and climate alignment. Saudi

Arabia can use its comparative advantages to become a leading,

cost-efﬁcient, and sustainable AI infrastructure hub in the region.

To achieve this balance, the analysis highlights four broad

areas to guide future policy and planning. Continued

investment in energy-efﬁcient computing technologies, such as

next-generation GPUs, advanced servers, and optimized cooling,

can increase computing output while managing power demand.

Developing AI-ready investment zones with reliable grid

connections and renewable integration could attract long-term

investors and strengthen Saudi Arabia’s position. Expanding

local research in data center efﬁciency, advanced cooling,

and sustainable design, together with collaboration between

universities, research institutions, and global technology

partners, would support knowledge transfer and industrial

diversiﬁcation. Operationally, emphasizing efﬁcient utilization

of new facilities, ﬂexible scheduling of AI workloads during

off-peak hours, and promoting resource reuse practices such as

heat recovery and water recycling could improve efﬁciency and

system integration. From a governance and energy planning

perspective, maintaining stable and transparent electricity

pricing, encouraging voluntary efﬁciency benchmarks, and

coordinating data center development with renewable energy

and grid expansion initiatives would align digital growth with

the Kingdom’s long-term energy transition objectives.

In conclusion, the Kingdom stands at a strategic inﬂection

point in the energy-digital nexus. Data centers are poised to

become a major new source of electricity demand, but their

role will depend on prudent planning, measured expansion,

and operational efﬁciency. A cautious, efﬁciency-oriented

approach – based on realistic demand assessment and strong

coordination with the national energy transition – will allow

Saudi Arabia to capture the long-term value of digital growth

while safeguarding reliability, affordability, and sustainability.

1 Introduction 09

2 Fundamentals of AI Data Centers 10

2.1 What Is an AI Data Center? 10

2.2 The Rising Demand for AI Data Centers 12

3 Overview of the Global AI Data Center Landscape 15

3.1 Power Capacity 15

3.2 Capital Investment 17

3.3 Electricity Demand 18

3.4 Emissions 19

3.5 Regional Outlook (Up to 2030) 20

4 Saudi Arabia’s AI Data Center Landscape 22

4.1 AI Data Centers: A Pivotal Shift 23

4.2 Demand Projections 25

4.3 Emissions Projections 27

4.4 Key Enablers 29

4.5 Factors that Could Inﬂuence Projections 30

5 Cost Analysis of AI Data Centers in Saudi Arabia 32

5.1 Calculating AI Data Center Project Costs 32

5.2 Baseline Results and Sensitivity Analysis 33

5.3 Policy Insights 38

6 Global Overview of AI Data Center Challenges and Risks 39

6.1 Challenges 39

6.2 Risks 40

7 Towards Sustainable and More Efﬁcient AI Data Centers 42

7.1 Techniques for Enhancing Operational Efﬁciency 42

7.2 Options for Decarbonization 45

7.3 Selected Regional Policies 46

Table of Contents

AI and Energy: The Future of Data Centers in Saudi Arabia

8 Conclusion and Recommendations 50

References 52

Acknowledgments 59

Appendices 60

Appendix A. Scenario-Based Framework for Estimating Data Center Capacity (2025–2030) 59

Appendix B. Methodology for Estimating Data Center Energy Demand in Saudi Arabia (2025–2030) 60

Appendix C. Methodology for Estimating CO2 Emissions from Data Centers’ Electricity Consumption 62

Appendix D. Methodology for Estimating Lifetime Data Center Project Costs 64

Appendix E. Glossary of Deﬁnitions 65

About the Authors 67

About the Project 69

Table of Contents

AI and Energy: The Future of Data Centers in Saudi Arabia 9

Data centers are at the heart of this shift. Operators are adding

more facilities and upgrading capabilities: clustering tens of

thousands of graphics processing units (GPUs), adopting high-

speed interconnects, deploying advanced thermal management,

and integrating tighter energy management and on-site power

options. AI-optimized hyperscale campuses are becoming the

main venues for model training and high-volume inference,

while enterprises modernize colocation facilities to keep

computing close to data, reducing movement and latency.

This rapid build-out has immediate energy implications.

AI workloads are more power-dense and often run at high

utilization, intensifying electricity demand and grid planning

needs. For instance, training GPT-4 is estimated to have used

approximately 42 gigawatt-hours (GWh) of electricity, and data

centers are projected to account for 3% of global power sector

demand and 1% of total energy sector emissions by 2030, with

AI-oriented servers expected to represent about half of the

sector’s electricity growth (Spencer et al. 2025). At the same

time, efﬁciency and sustainability strategies, such as efﬁcient

Artiﬁcial intelligence (AI) has grown explosively in recent years, becoming a driving force

across industries and economies worldwide. AI adoption is accelerating with large language

models (LLMs), generative AI (GenAI) applications, and automation in manufacturing and

services. As models grow more capable and are deployed at massive user scale, the need

for high-performance computing accelerates. Training and serving today’s frontier models

increasingly rely on specialized accelerators, high-bandwidth networks, and dense storage,

pushing rapid evolution in the digital infrastructure that underpins the AI economy.

Introduction

use, workload scheduling, renewable energy, and emerging low-

carbon baseload options, are becoming central to efﬁcient data

centers. The balance between growth and sustainability will

shape the AI sector’s trajectory and national energy systems.

This study examines the future of AI data centers in Saudi

Arabia, beginning with an overview of AI data center

fundamentals, global trends shaping their expansion,

investment patterns, electricity demand, and sustainability

challenges. It then analyzes the current landscape in the

Kingdom and outlines potential growth scenarios for AI-

oriented centers. A detailed cost and efﬁciency analysis

assesses competitiveness under different electricity pricing

and infrastructure conditions. The study also discusses

the main risks associated with rapid expansion, proposes

mitigation measures, and reviews international approaches

to developing sustainable AI infrastructure. It concludes with

recommendations to support Saudi Arabia’s national objectives

while strengthening its position in the global data center

economy.

2.1 What Is an AI Data Center?

An AI data center is a specialized facility designed to handle

the intensive computational demands of AI workloads,

including training, deploying, and running AI applications and

services. These data centers feature advanced computing,

networking, and storage infrastructures, along with robust

energy and cooling capabilities, to effectively run AI algorithms

at scale (IBM 2025). Unlike traditional data centers, which

support general IT operations such as web hosting, enterprise

applications, and data storage, AI data centers are purpose-built

for high-performance computing. While traditional and AI data

centers share core components such as servers, storage, and

security measures, as shown in Figure 1, AI data centers differ

signiﬁcantly in the scale and capability of their hardware and

architecture.

In other words, AI data centers are optimized environments

capable of managing complex AI models with billions of

parameters and delivering real-time AI inference across a wide

range of applications. Table 1 summarizes the differences

between traditional and AI data centers.

This chapter introduces the fundamentals of AI data centers, highlighting how they differ

from traditional facilities in design, performance, and purpose. It reviews the main types

of AI data centers currently in use and explains the rapid growth in demand driven by AI

applications. The discussion concludes by examining the link between AI data centers and

their rising energy needs.

Fundamentals of AI

Data Centers

AI data centers come in several forms, reﬂecting different

scales, ownership models, and deployment conﬁgurations. The

main categories of data centers supporting AI growth today are

outlined below, and Table 2 shows a comparative summary of

the three types (IBM 2025; Shehabi et al. 2024; Duncan et al.

2024):

• Hyperscale AI data centers: Ultra-large data centers

typically housing 5,000 servers and spanning at least

10,000 sq ft.1 They are engineered for extreme scalability

and designed to handle large-scale AI workloads, including

training frontier AI models and providing AI services

to millions of users. These facilities are often built and

operated by major cloud providers, such as Amazon Web

Services, Google Cloud, Microsoft Azure, and Meta.

• Colocation AI data centers: Large data centers owned

and operated by third-party providers who rent out space,

power, and network connectivity to other companies. A

provider can deploy or rent ready-to-use servers inside

these colocation facilities. This allows businesses of all

sizes to access AI-grade infrastructure without the costs of

building and maintaining their own.

1 Sq ft = Square feet.

AI and Energy: The Future of Data Centers in Saudi Arabia

Figure 1. Data center’s components.

Note: Schematic of a traditional data center. Network connections link the facility to external digital infrastructure, enabling continuous data exchange. The light green elements

illustrate data ﬂow between internal servers and outside networks. The dark green elements indicate electricity supply and cooling systems that maintain stable operating

conditions. UPS provides short-term backup power, ensuring continuous operation during grid ﬂuctuations or outages.

Source: IEA (2025). Reproduced by authors for improved readability.

IT equipment housed in racks

Servers Storage systems

Networking equipment

UPS (uninterruptible

power supply)

Grid

connection Backup

generator

Network connections

Cooling

Table 1. Differences between traditional and AI data centers.

Feature Traditional data centers AI data centers

Workload type General computing, storage, web hosting,

and enterprise IT AI training and inference, and high-performance computing (HPC)

Hardware focus Central processing units (CPUs) and modest accelerator

use

GPUs, tensor processing units (TPUs), or AI accelerators (e.g.,

NVIDIA’s H100 or GB200)

Cooling technology Air cooling Advanced liquid cooling, direct-to-chip or immersive cooling

Storage architecture Optimized for structured storage: block-level block-level

SAN2 and relational database systems

Advanced, scalable, high-throughput storage systems (such as

parallel ﬁle systems, NVMe-based,3 or object storage)

Power density 5-8 kilowatts per rack 30+ kilowatts per rack

Networking Ethernet with latency-tolerant topologies High-bandwidth, low-latency networking

Scalability Moderate: scale by adding standalone halls or cages High scalability for AI clusters

Facility architecture Single-story or low-rise buildings, standard 3-4m ceiling,

raised-ﬂoor or slab

High-bay (6-10m) or multi-story blocks with reinforced ﬂoor loads,

overhead liquid manifolds, and busways

Power capacity 10-30 MW 100-1,000+ MW

2 SANs: Storage area networks.

3 NVMe: Non-Volatile Memory Express.

AI and Energy: The Future of Data Centers in Saudi Arabia

• Enterprise AI data centers: Facilities owned and operated

by a single organization for its own use – typically on their

premises or at company-controlled sites. Enterprises

maintain these facilities to retain control over sensitive data

and comply with data sovereignty concerns. These centers

are used by organizations with particularly sensitive data,

latency-critical applications, or large-scale internal needs to

deploy AI-ready infrastructure on the premises.

2.2 The Rising Demand for AI

Data Centers

AI development and adoption have grown exponentially in

recent years, driven by advances in machine learning, especially

deep learning, and the explosion of big data. AI is no longer

limited to research labs; it has become a key driver of economic

growth, boosting productivity and strengthening both global

and national competitiveness. According to the International

Data Corporation (IDC), the global AI market was valued at

nearly $235 billion in 2024 and is projected to reach almost

$631 billion by 2028 (International Data Corporation 2024).

Additionally, AI could contribute about $19.9 trillion to the

global economy by 2030 and drive 3.5% of global GDP in that

year (Fioretti et al. 2024).

A major force behind this rapid growth is GenAI and LLMs. These

technologies have made AI more accessible, leading to record-

breaking adoption rates across businesses worldwide. McKinsey

reports that in 2025, nearly 78% of companies are using AI in

at least one area of their operations, with 71% using GenAI

speciﬁcally, a rate that has more than doubled from just 33%

in 2023 (Singla et al. 2025). This clear potential for productivity

and innovation has led to an explosion of AI applications, from

AI copilots in software to AI-driven analytics, across sectors such

as ﬁnance, healthcare, education, and manufacturing.

With this expansion, the need for robust computation

resources has risen to support AI innovation. Between 2012

and 2018, the compute required for leading AI training runs

doubled roughly every 3-4 months (Lohn and Musser 2022),

and although advances in algorithms and architecture have

improved efﬁciency, the emergence of LLMs has kept the

demand for computing growing at a similar rate. As a result,

organizations across industries, including banking, health care,

and manufacturing, are investing heavily in advanced computing

infrastructure to support AI development. The IDC projects

that AI will continue to drive new data center investments,

with spending expected to grow at a compound annual growth

rate (CAGR) of 36% between 2022 and 2027 (Hoff 2024), as

Figure 2 shows.

Table 2. Differences between AI data centers.

Type Hyperscale Colocation Enterprise

Ownership and use Cloud giants Owned by a third party and used by multiple tenants Single organization

Floor space (sq ft) 10,000-1,000,000+ Varies 5,000-20,000

Server count 5,000+ Varies 500-2,000

Power demand (MW) 20-100+ Varies 5-10

AI and Energy: The Future of Data Centers in Saudi Arabia

The accelerated adoption of AI and the rise of compute-

intensive workloads are reshaping data center infrastructure

requirements: how they are designed, how many are needed,

and how large they must be. This transformation brings both

opportunities and challenges for investors and policymakers.

Central to these challenges are the signiﬁcant power and

environmental implications. High-density servers running

continuous training cycles and large-scale inference workloads

consume vast amounts of electricity. In some cases, a single

hyperscale AI campus may use as much power as a small city.

Figure 3 shows that energy requirements vary across different

stages of the AI lifecycle, and consistently exceed those of

traditional digital services. Early estimates indicate that a

generative AI query consumes roughly 10 times more electricity

than a conventional search, rising from about 0.3 watt-hours

(Wh) for a Google search to nearly 2.9 Wh for a ChatGPT request

(EPRI 2024). At scale, such differences can strain local grids.

Regions with clusters of data centers are already facing capacity

constraints, proving the urgency of strategic planning.

Understanding the current state, trajectory, and needs of AI

data centers has become an economic, environmental, and

geopolitical necessity. AI data centers create opportunities

for investment, innovation, and digital competitiveness, but

they also pose challenges in securing reliable electricity and

maintaining sustainability. Nations that can effectively align

AI infrastructure growth with energy system resilience will be

better positioned to reap the beneﬁts of the AI economy.

Figure 2. Growth in data center spending due to AI between 2022-2027.

Source: Hoff (2024). Reproduced by authors for improved readability.

AI infrastructure semiconductors ($M) Non-AI infrastructure semiconductors ($M)

$300,000

$250,000

$200,000

$150,000

$100,000

$50,000

2022 2023 2024 2025 2026 2027

Strong Growth in AI Spending

36% 5yr CAGR

AI and Energy: The Future of Data Centers in Saudi Arabia

For Saudi Arabia, understanding and leveraging these

dynamics is essential to attract AI investment, expand digital

infrastructure, and ensure a reliable energy supply. Effectively

balancing digital expansion with energy transformation will

be critical to securing a strategic role in the global AI economy

and advancing national objectives for economic diversiﬁcation,

such as R&D priorities, telecommunications, IT, and industrial

sector goals.

Figure 3. Energy consumption across AI lifecycle.

Source: Greene-Dewasmes, Higgins, and Tladi (2025). Reproduced by authors for improved readability.

Stage 5:

Monitoring and

maintenance

Stage 4:

Deployment

60%

Stage 3:

Model training

30%

Stage 2:

Model development

10%

Stage 1:

Planning and data

collections on nature

3.1 Power Capacity

The global data center landscape is expanding rapidly in both

number and power capacity. As of March 2024, more than

11,800 facilities were operational worldwide, about twice as

many as ﬁve years earlier. The U.S. leads with 5,388 data centers

(around 45% of the total), followed by Germany (520), the

United Kingdom (512), China (449), and Canada (336) (Fleck

2024). Within this broader build-out, hyperscale platforms that

power AI workloads and the cloud had surged

The rise of AI is becoming the dominant driver of demand in the global data center

industry, reshaping infrastructure growth, investment ﬂows, energy consumption, and

environmental footprints. This section outlines the state of AI data centers worldwide and

evaluates their implications for infrastructure, energy demand, and economics.

Overview of the

Global AI Data

Center Landscape

to about 1,189 data centers by early 2025 (Synergy Research

Group 2025). According to S&P Global Data, which we primarily

use for our analysis and projections throughout this paper

(unless otherwise stated), the global installed capacity

of data centers has been estimated to be around 111,897

MW in 2024, as shown in Figure 4. While enterprise data center

capacity has remained relatively stable, signiﬁcant expansion

is driven by hyperscale and colocation facilities, which

approximately account for 15.6% and 44.9% of the

global capacity, respectively, since 2020.

AI and Energy: The Future of Data Centers in Saudi Arabia

The distribution of the overall capacity is uneven around the

globe. In fact, by 2024, the U.S. and China account for roughly

60% of this capacity (38,962 MW and 28,639 MW, respectively),

while the European Union holds about 20,093 MW, as

Figure 5 shows.

Figure 5. Global distribution of installed data center capacity in 2025.

4500–9000

700–1150

125–200

Planned Current

Source: Authors.

Figure 4. Historical and projected global installed data center capacity (MW).

Source: Authors.

Historical Projected

250,000

200,000

150,000

100,000

50,000

2019 2020 2021 2022 2023 2024 2025E 2026E 2027E 2028E 2029E 2030E

Enterprise Hyperscale Colocation

AI and Energy: The Future of Data Centers in Saudi Arabia

Looking ahead, S&P Global estimated that by 2030, the total

installed capacity in data centers will be around 224,360 MW,

as Figure 4 shows, with a CAGR of 12% between 2024 and 2030.

Hyperscale and colocation data centers are expected to grow

by a CAGR of 17%, and enterprise centers by only 5% . For

AI-speciﬁc IT capacity, the IDC expected an increase to 19,600

MW by 2027, up from 3,600 MW in 2022 (Graham, Rutten, and

Yashkova 2024). McKinsey, on the other hand, predicted that in

a midrange scenario, the demand for AI data center capacity will

rise to 33% on average between 2023 and 2030, meaning that by

2030, advanced AI-equipped facilities are expected to account

for about 70% of total data center capacity, with generative AI

workloads representing roughly 40% of that share (Srivathsan

et al. 2024). All these factors underscore the industry’s shift

toward high-density, AI-optimized compute.

3.2 Capital Investment

The rapid growth of AI has triggered a global wave of capital

investment in data center infrastructure. Companies and

governments are allocating funds to expand data center

capacity through new buildings (greenﬁeld) and upgrades

of existing facilities (brownﬁeld) to meet the demand for AI

training and inference workloads.

According to CBRE Group’s4 global survey, 95% of 92 major data

center investors planned to increase spending in 2025, with

41% expecting to allocate $500 million to $2 billion (or more),

up from 30% in 2024 (CBRE 2025). The IEA reported that global

investment in data centers increased by nearly 70% in the last

two years (Spencer et al. 2025).

Many governments have launched initiatives to build data

centers. For example, the U.S. launched the Stargate AI initiative

in 2025 with a $500 billion investment to run through 2028.

The European Union launched InvestAI, dedicating $207 billion

to build “AI gigafactories” that provide critical computational

infrastructure, including data centers (Clemmons and Graham

2025). China invested over $6.1 billion in 2024 to build eight

computing hubs as part of its mega data initiative “East Data,

West Computing.” This led to additional data center investment

from provincial governments and state-afﬁliated ﬁrms, totaling

$27 billion by the end of the same year to further boost AI

computing capacity (Stokols 2025). In parallel, France and the

UAE jointly announced a plan to build a 1,000 MW AI data center

in Europe, with an estimated investment of $34.8 to $58billion

(Narasimhan et al. 2025).

Hyperscale cloud providers remain the primary drivers of AI

infrastructure investment. Google, Amazon, Meta, and Microsoft

collectively planned to spend $320 billion in 2025, up from $230

billion in 2024, representing a 1.4-times increase and a CAGR of

40.37%. Microsoft alone committed $80 billion for AI-enabled

data centers, while Meta projected $60 to $65 billion for AI

supercomputing (Houlihan Lokey 2025), as shown in Figure 6.

If current trends continue, hyperscale capital expenditures

(CAPEX) are expected to push annual global data center

investment above $1 trillion by 2029, more than double

today’s level. McKinsey estimated the investment needed

worldwide in data centers to meet AI demand alone

to be $5.2 trillion, and it’s projected to reach nearly $7 trillion

by 2030 (Noffsinger et al. 2025). This increase is driven by

AI-driven demand for advanced chips, cooling systems,

and grid integration, as well as power and land constraints.

4 CBRE Group: Coldwell Banker Richard Ellis.

AI and Energy: The Future of Data Centers in Saudi Arabia

3.3 Electricity Demand

Since 2017, data center electricity consumption has

accelerated due to technological trends such as the growth

in cloud computing, the use of social media, and the rapid

expansion of AI. According to the IEA, the yearly rate of data

centers’ electricity consumption grew from 3% between 2005

and 2015 to 10% between 2015 and 2024 (Spencer et al. 2025).

In fact, S&P Global estimated the global data center electricity

consumption to be around 854 TWh.

AI workloads are the most signiﬁcant driver of this surge.

Studies estimated that AI-speciﬁc tasks accounted for only 5%-

15% of global data center energy use in 2023, but are projected

to keep increasing rapidly to 35%-50% in 2030 (Kamiya and

Coroamă 2025). Accordingly, the global electricity demand in

data centers will double by 2030 to about 1,891 TWh with a

CAGR of 14%,5 as shown in Figure 7.

Despite these ﬁgures, the outlook is far from certain, and it

is plausible that such projections may not materialize. The

trajectory will depend on several evolving factors, including

the pace of AI deployment, regulatory developments, and the

readiness of power and grid infrastructure. In many markets,

delays in securing grid connections and long lead times for high-

capacity transformers have become critical bottlenecks, often

extending project timelines by up to three years.

These constraints, coupled with rising costs, could slow the

build-out of large-scale AI campuses and temper near-term

growth expectations. For example, the IDC anticipated an

AI-driven increase in data center electricity use at a CAGR of

44.7%, reaching about 146.2 TWh by 2027 (Graham, Rutten,

and Yashkova 2024), while Goldman Sachs projects AI-related

power use could grow to 200 TWh per year between 2023 and

2030 (Goldman Sachs 2024). Even with more conservative

growth, the expanding digital infrastructure will have signiﬁcant

implications for the environment and energy systems,

highlighting the importance of efﬁciency and sustainability

measures in the sector’s evolution.

Figure 6. Hyperscale cloud providers actual and projected spending on AI data center development in 2023-2025.

Source: Houlihan Lokey (2025). Reproduced by authors for improved readability.

$0.00

$20.00

$40.00

$48.40

$31.90 $32.30

$52.50

$28.10

$39.20

$60.00

$80.00

$77.00

$100.00

$60.00

$80.00

$100.00

$120.00

2023 ($B) 2024 ($B) 2025E ($B)

$55.70

$75.00

Microsoft

5 S&P Global dataset.

AI and Energy: The Future of Data Centers in Saudi Arabia

Figure 7. Historical and projected global data center energy consumption (TWh) for 2019-2030.

3.4 Emissions

The energy surge from AI data centers also carries a signiﬁcant

environmental footprint. Its carbon emissions could challenge

global climate targets without proper mitigation. The IEA

estimates that data centers currently account for around 180

Mt CO2 globally, less than 1% of total energy-related emissions,

yet they are among the fastest-growing sources of emissions

in the power sector, projected to peak at about 320 Mt CO2 by

2030 (Spencer et al. 2025). Accenture estimates that by 2030,

emissions from AI data centers may reach 3.4% of worldwide

greenhouse gases (CO2e), a roughly 11-fold increase within

a decade (Jamison et al. 2025). In this study, emissions are

reported in CO2 from fossil-fuel-based electricity generation and

exclude other greenhouse gases. AI infrastructure could emit

300-350 Mt CO2 annually by 2030, comparable to the emissions

of a mid-sized country.

Most major hyperscale operators have pledged to achieve net-

zero emissions, adding more complexity to the sector’s energy

strategy. Meeting these targets requires a growing share of

renewable electricity, yet those sources are intermittent. As a

result, data center operators increasingly rely on natural gas or

nuclear generation for reliable power while offsetting emissions

through renewable energy certiﬁcates (RECs). Developing

reliable REC markets, whether domestic or international,

is therefore critical, and their current immaturity could

delay new projects or raise compliance costs. Alternatively,

operators may look for more ﬂexible emission targets or slower

implementation timelines to balance growth with sustainability

commitments.

Beyond carbon, other environmental pressures are rising,

notably water use and heat emissions. Most AI data centers rely

on power-hungry cooling systems, such as chilled water and air

conditioning, to remove the signiﬁcant heat generated by

high-performance chips. By one estimate, cooling AI data

centers in 2030 will require about 3.02 billion cubic meters

of fresh water annually, more than countries like Norway or

Sweden withdraw in a year (Jamison et al. 2025). Even a small-

scale data center can have a substantial water footprint, where

every 1 MW of data center capacity demands 25-30 million liters

of water per year for cooling, equivalent to the drinking water

needs of about 300,000 people (Spindler, Hahn-Petersen, and

Hosseini 2024).

Source: Authors.

Historical Projected

Enterprise Hyperscale Colocation

2,000

1,800

1,600

1,400

1,200

1,000

800

600

200

2019 2020 2021 2022 2023 2024 2025E 2026E 2027E 2028E 2029E 2030E

400

AI and Energy: The Future of Data Centers in Saudi Arabia

3.5 Regional Outlook

(Up to 2030)

This section provides the current state and projected outlook

across major geographies, highlighting demand for data

center capacity, investment hotspots, and growth in electricity

consumption.

North America (United States)

North America, particularly the U.S., leads the world in

data center capacity and AI compute, hosting the largest

concentration of hyperscale and AI facilities. U.S. cloud

providers serve as the primary backend for the world’s AI

activity, with a signiﬁcant portion of inference and training tasks

performed globally being executed on servers in U.S.-operated

data centers. Major hubs include the West Coast, Texas, and

Virginia, with Northern Virginia, often referred to as “Data

Center Alley,” housing the single largest concentration of cloud

and AI infrastructure in the world (Spencer et al. 2025).

According to S&P Global, the total installed data center capacity

in the U.S. is projected to reach about 89,560 MW by 2030. This

expansion has signiﬁcant energy implications. U.S. data centers

consumed roughly 322 TWh of electricity in 2024, as Figure 8

shows. This started in 2017, after a period of stability in annual

energy consumption between 2014 and 2016 at about 60 TWh,

driven by the growth in the installed server base and GPU-

accelerated servers for AI (Shehabi et al. 2024). Looking ahead,

energy consumption is expected to increase to about 769 TWh

by 2030 (Green et al. 2024).

Figure 8. Historical and projected regional data center electricity consumption (TWh) for 2019-2030.

Note: APAC: Asia-Paciﬁc, MEA: Middle East and Africa, LATAM: Latin America.

Source: Authors.

2019

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2020 2021 2022 2023 2024 2025E 2026E 2027E 2028E 2029E 2030E

North America (U.S.) Europe China APAC (exc. China) MEA LATAM

Historical Projected

AI and Energy: The Future of Data Centers in Saudi Arabia

China

China is one of the world’s major players in AI data center

infrastructure. It hosts some of the largest data center

installations and is racing to expand its AI compute capacity.

Over the past decade, it has built a vast domestic network

of hyperscale facilities supporting tech giants like Alibaba,

Tencent, and Baidu. There are also state-led AI initiatives. In

2024, China hosted about 28,639 MW of data center capacity,

which is expected to increase to more than 54,614 MW in 2030.

China currently represents about 25% of global data center

electricity consumption (Spencer et al. 2025), about 201 TWh

of electricity. The data center sector in China started to expand

in 2015, with electricity demand growing by 15% per year until

2024 (Spencer et al. 2025). This growth is expected to continue

to 2030, with demand projected to reach more than double, as

the S&P Global data show.

Europe

Europe is expanding its AI data center footprint. While

historically behind the U.S. in hyperscale cloud, the region is

now investing in its own AI compute and regional capacity.

Major hubs include Dublin, Frankfurt, Amsterdam, London, and

the Nordic countries. In 2024, the total installed capacity in all

data centers within the European region reached around 20,000

MW, expected to increase to approximately 33,422 MW by 2030.6

Germany led with 3,188 MW in 2024, followed by the UK with

2,764 MW and France with 1,374 MW.

This trend stems from the EU’s ambition to become a global

leader in AI by building AI factories and supercomputing centers

across member states, with plans to triple data center capacity

in the next ﬁve to seven years (European Commission 2025).

Data centers consumed about 151 TWh of electricity in 2024,

and this is estimated to grow by more than 281 TWh in 2030,

with a CAGR of 11%. Also, a notable trend in the region is the

rise of data center campuses in the Nordic countries speciﬁcally

targeting AI customers with 100% green energy and high-

performance computing offerings (Duncan et al. 2024).

GCC and the Middle East

The Middle East is rapidly becoming a global hotspot for data

centers and AI compute, driven by economic diversiﬁcation

strategies and favorable energy conditions. The Gulf

Cooperation Council (GCC) countries are leading this trend,

capitalizing on low-cost land and energy; their strategic location

bridging Asia, Africa, and Europe; and available capital. Despite

the region’s hot climate, which increases cooling needs,

governments see data centers as key pillars of their digital and

economic transformation agendas. National strategies, such as

Saudi Arabia’s Vision 2030, UAE Vision 2031, New Kuwait 2035,

and Digital Oman 2030, have boosted the Middle East’s data

center market (Tohme et al. 2025).

As of 2024, the region hosted more than 290 data centers in

17 countries (Data Center Map.n.d.). Total data center capacity

reached about 5,311 MW in 2024 and is expected to grow to

around 8,938 MW in 2030. Industry analysts have stated that

the Middle East is one of the fastest-growing regions for data

centers (Shiwani, Abbasi, and Levack 2025).

Flagship projects illustrate this trend. In Saudi Arabia, major

initiatives such as HUMAIN’s AI factories in Dammam and

NEOM’s Oxagon Green AI Campus are driving the region’s

growth, with multi-gigawatt capacity planned by 2030. UAE has

partnered with U.S. ﬁrms to develop a 5,000MW AI data center

campus through G42, aiming to build the largest AI-dedicated

data center complex outside the U.S. The ﬁrst 1,000 MW phase

is already in progress (U.S. Department of Commerce 2025).

Qatar, Oman, Bahrain, and Kuwait are also expanding. Oman,

for example, has a partnership with Equinix for a large hub in

Salalah, and Kuwait has announced a 1,000 MW data center park

(Shiwani, Abbasi, and Levack 2025). Beyond the Gulf, secondary

markets like Egypt and Morocco are investing in new data center

parks and incentives to attract cloud providers. S&P Global data

project that energy consumption of data centers in the Middle

East will rise from 38 TWh in 2024, with a CAGR of 11%, to

around 72 TWh in 2030.

The global AI data center landscape is shifting from a U.S.- and

China-centered model toward a more distributed network of

regional centers. As countries increasingly view AI compute

as a strategic asset, regions are capitalizing on their distinct

advantages to build competitive AI capabilities: the U.S. tech

ecosystem, China’s scale and state support, Europe’s regulatory

environment and green power, and the Middle East’s energy

resources. By 2030, a more balanced global distribution is likely

to have materialized, with emerging regions contributing more

substantially to AI workloads.

6 S&P Global data.

The development of data centers in Saudi Arabia has historically

aligned with the Kingdom’s broader digital transformation

strategies. In the early 2000s, infrastructure was primarily built

by national telecom operators to support government agencies,

national databases, and basic internet services. As demand for

digital services grew, the 2010s saw an expansion in colocation

and enterprise-grade facilities driven by the ﬁnancial sector,

e-government initiatives, and cloud adoption. The launch of

Vision 2030 in 2016 marked a turning point, positioning digital

infrastructure as a cornerstone of economic diversiﬁcation.

Since then, the sector has evolved from general-purpose

hosting to advanced, AI-capable infrastructure as part of the

Kingdom’s digital economy goals.

As of 2024, Saudi Arabia has around 58 data centers (CST 2024a).

Between 2022 and 2024, the IT capacity of these data centers

more than doubled, from 122.4 MW (MCIT 2023) to 290.5 MW

(MCIT 2024), reﬂecting a CAGR of approximately 54% and placing

the Kingdom among the fastest-growing digital infrastructure

Saudi Arabia is becoming one of the most dynamic data center markets in the Middle

East, with AI-driven infrastructure a key pillar of its digital and energy transformation.

Rapid growth in the sector reﬂects the Kingdom’s national digital strategies, supported

by favorable regulation, a reliable power supply, and strong government and private

investment. This section examines the evolution and distribution of data centers across

the country, the factors enabling their growth, and the implications for electricity demand

and emissions. It also develops a set of growth scenarios for the period 2025 to 2030

to assess how different operational and efﬁciency conditions could shape the sector’s

contribution to Saudi Arabia’s energy and sustainability goals.

Saudi Arabia’s

AI Data Center

Landscape

markets in the Middle East. Saudi Arabia’s data centers are

concentrated in a few key regions. According to S&P Global,

approximately 79% of the country’s installed capacity is located

in two cities: Riyadh and Dammam, as shown in Figure 9.

This spatial concentration reﬂects the country’s economic

geography: the availability of critical infrastructure, such

as power grids and ﬁber-optic networks, and the proximity

to enterprise markets. Riyadh, with nearly 38% of the total

capacity, has become the main hub for general-purpose cloud

and enterprise data centers, as it hosts many government

agencies and company headquarters. Dammam accounts for

the largest share at roughly 41% of capacity. Its location in

the Gulf industrial corridor, close to subsea cables and major

energy companies, drives its rapid growth as a fast-emerging

computing zone. Jeddah contributes about 7% of the capacity

and other cities, including Madinah and Buraydah, each hold

only 1%-3% of the total capacity; these smaller facilities mainly

serve local government and business needs.

AI and Energy: The Future of Data Centers in Saudi Arabia

4.1 AI Data Centers: A Pivotal

Shift

Saudi Arabia’s data center capacity is growing fast, mostly

from AI data centers. Until 2024, most facilities were primarily

designed to support cloud services, enterprise hosting, and

digital government services. Starting in 2025, there has been a

pivotal shift toward AI-oriented infrastructure.

The ﬁrst clear step in this direction came with the early 2025

launch of Groq Cloud’s inference-ﬁrst facility in Dammam,

signaling the Kingdom’s shift in the data center landscape as

the facility is explicitly built for AI at scale (Groq 2025). At the

same time, HUMAIN’s ﬁrst 50 MW phase is scheduled to go live

late in 2025, with further expansion planned throughout the

decade (Techusiness 2025). NEOM is also developing the

Oxagon AI Campus, starting with 300 MW in 2028 (NEOM 2025).

By the early 2030s, projects from HUMAIN, NEOM, and

others are expected to push national capacity into the

multi-gigawatt range.

Figure 9. Data center capacity distribution across Saudi Arabia regions.

14%

41%

38%

Riyadh Dammam Jeddah Other

By mid-2025, announced investments in Saudi Arabia’s data

center sector had surpassed $25 billion, with long-term

commitments projected to reach about $77 billion by 2034 (Saudi

Market Research Consulting Firm 2025). HUMAIN is leading much

of this growth, working with international technology partners

like NVIDIA, AMD, Amazon Web Services (AWS), Qualcomm,

and Cisco. NEOM’s AI campus is another major contributor. If

all announced projects go ahead, Saudi Arabia could account

for up to 7% of global AI training and inference capacity by the

early 2030s (England and Omran 2025). Notable key projects as

of mid-2025 are summarized in Table 3. This transition marks

a shift from earlier HPC systems, such as KAUST’s Shaheen

and Aramco’s Dammam-7, toward large-scale, commercial AI

infrastructure supporting national and global digital economic

growth.

Current and announced data centers and their capacity will be

distributed throughout the Kingdom. Projects will include scaling

existing hubs, such as Riyadh and Dammam, and creating new

ones, such as NEOM in the northwest, as shown in Figure 10.

However, some announced capacities have not yet been assigned

to speciﬁc locations.

Source: Authors.

AI and Energy: The Future of Data Centers in Saudi Arabia

Table 3. Major Saudi Arabia AI data center projects based on public announcements (as of mid-2025).

Project/ initiative Timeline Lead entity Investment Capacity target Key objectives

HUMAIN-Groq AI Inference

Center (Groq 2025)

2024-2025 HUMAIN and Groq $1.5 B N/A Inference-only facility using Groq’s LPUs,

optimized for ultra-low-latency generative AI

workloads

Oracle Infrastructure

Investment (Egbert 2025)

2024-2034 Oracle $14 B N/A Expansion of cloud and AI infrastructure to

support national digital transformation and

enterprise services

HUMAIN-AWS AI Zone

(Amazon 2025)

2025-2026 HUMAIN and AWS $5 B N/A Dedicated AI innovation zone integrated with

AWS infrastructure

Google Cloud-PIF AI Hub

(Google Cloud 2025; PIF 2024)

2025-2027 HUMAIN and Google

Cloud

$10 B N/A Regional hub for AI development and cloud

services, with a focus on enterprise and

government applications

HUMAIN-AMD AI

Infrastructure (Techusiness

2025; HUMAIN 2025; Grabein

and Stine 2025)

2025-2030 HUMAIN, AMD, and

Cisco

$10 B 500 MW (part of

1,900 MW by 2030;

6,600 MW by 2034)

Deploy AMD MI300 accelerators with Cisco

networking to build open infrastructure

targeting developers, startups, and national

LLM projects

HUMAIN-NVIDIA AI Factories

(NVIDIA 2025a)

2025-2030 HUMAIN and NVIDIA $10 B 500 MW (part of

1,900 MW by 2030;

6,600 MW by 2034)

Initial 50 MW deployment with 18,000 NVIDIA

GB300 GPUs, phased expansion to 500 MW

with 180,000 GPUs

NEOM-DataVolt Oxagon

Green AI Campus insert

space (NEOM 2025)

2025-2031 NEOM and DataVolt $5 B 1,500 MW Net-zero hyperscale AI campus in Oxagon

storage, and hydrogen backup with advanced

cooling systems

HUMAIN-Qualcomm Hybrid

Inference (Qualcomm 2025)

2026-2028 HUMAIN and

Qualcomm

$2 B N/A Cloud-to-edge hybrid AI ecosystem, focusing

on energy-efﬁcient, low-latency inference for

mobile and embedded systems

Figure 10. Current (2024) and planned (2030) data center hub capacities in Saudi Arabia.

Note:Planned capacity (1,716 MW) only includes projects with speciﬁed locations.

Source: Authors.

700 MW

200 MW

5 MW Planned Current

Madinah

Jeddah

Neom

Buraydah Riyadh

Dammam

Alkhobar

AI and Energy: The Future of Data Centers in Saudi Arabia

This expansion is expected to have signiﬁcant implications for

the Kingdom’s digital infrastructure and energy landscape.

Large AI campuses will place high-density demands on Saudi

Arabia’s electricity system in the coming years. AI introduces

a new type of workload, namely continuous, GPU-intensive

operations that differ from traditional data center activity.

Understanding the future impacts of this shift on grid stability,

energy sourcing, and sustainability will be critical for planning

and coordinating the next phase of infrastructure growth.

4.2 Demand Projections

Understanding the future impact of data centers requires

analyzing both the scale of potential capacity expansion and

the operational conditions of facilities. While numerous projects

have been announced, their ultimate realization, scale, and pace

of deployment are uncertain. To account for this uncertainty,

three growth scenarios for total installed capacity by 2030 are

deﬁned, as shown in Figure 11:

• Baseline scenario: This scenario reflects the continuation

of the current trends, with growth driven primarily by

general-purpose data centers, without a major expansion

of AI-related centers. This serves as a reference point

for capacity growth in the absence of large-scale AI

transformation.

• Moderate-growth scenario: This looks at confirmed

AI-oriented expansion, consisting of projects that are in

the stages of permitting or early execution, based on an

analysis of multiple data sources. This path is a realistic

and achievable AI growth trajectory, with total capacity,

combined with general-purpose data center expansion,

potentially reaching 2,000 MW by 2030.

• High-growth scenario: This represents visionary growth

in which all announced AI mega-projects are completed

by 2030, adding roughly 3,050 MW of new AI-optimized

capacity. Combined with the development of general-

purpose data centers, this brings total installed capacity

to around 4,100 MW by 2030. While these projects are

currently under construction, they set the upper limit of

potential capacity at the end of the decade.

Figure 11. Historical and projected Saudi Arabia data center capacities (MW) under three growth scenarios 2022-2030.

Note: Historical capacity is shown up to 2024 (290.5 MW). The non-AI scenario reaches about 1,050 MW by 2030, moderate growth reaches 2,000 MW, and high growth reaches

4,100 MW. The gray dashed line shows the national target of 1,300 MW.

Source: Authors.

ProjectedHistorical

290.5 MW

1050 MW

2000 MW

4100 MW

500

1,000

1,500

2,000

2022 2023 2024 2025E 2026E 2027E 2028E 2029E 2030E

2,500

3,000

3,500

4,000

4,500

Historical High Moderate Non-AI

National Target(1300 MW)

AI and Energy: The Future of Data Centers in Saudi Arabia

Table 4. Modeled scenarios for data center capacity growth and operational conditions in Saudi Arabia through 2030.

Operational condition

Conventional

(PUE 1.5-1.7, as-is emissions)

Sustainable

(PUE 1.3-1.5, low-carbon mix)

Growth

scenario

High growth

(4,100 MW by 2030)

S1: High growth, conventional

• AI-ﬁrst expansion

• High electricity use and as-is CO2 emissions

S2: High growth, sustainable

• Green AI expansion

• Improved efﬁciency from advanced cooling and chip design

• Lower electricity use and CO2 emissions

Moderate growth

(2,000 MW by 2030)

S3: Moderate growth, conventional

• Steady AI expansion

• Limited efﬁciency gains

• Higher electricity use and CO2 emissions

S4: Moderate growth, sustainable

• Balanced AI growth

• Enhanced efﬁciency from advanced cooling and hardware

optimization

• Lower electricity use and CO2 emissions

While these scenarios outline how far Saudi Arabia’s data-center

capacity could expand by 2030, their broader impact depends on

the efﬁciency of facilities and the sustainability of their energy

supply. To explore these dynamics, two operational conditions

are deﬁned:

• Conventional operational conditions: Efficiency

improvements are limited, advanced cooling technologies

are slow to be adopted, and most electricity comes from

fossil sources, typically resulting in higher power usage

effectiveness (PUE)7 values.

• Sustainable operational conditions: High operational

efficiency achieved through advanced cooling, optimization

technologies, and energy management practices.

A significant proportion of electricity comes from renewable

sources, resulting in lower PUE and emissions.

Combining moderate- and high-growth scenarios with these

two operational conditions yields four possible combinations,

shown in Table 4. These scenarios offer a unique perspective

for evaluating the long-term impacts of AI-driven data

center growth on the power system and sustainability

goals. Importantly, these scenarios are not predictions

but structured “what-if” analyses to stress-test policy and

investment choices.

As data center capacity expands, translating installed IT

load into electricity consumption is essential for assessing

implications for the national power system. The most widely

used methodology is a bottom-up approach (Masanet, Lei,

and Koomey 2024), which calculates electricity use based on

the power draw of IT equipment. This method provides the

most accurate estimates but relies on detailed technical data,

which is often unavailable for projects still in the planning

stages. In such cases, electricity demand can be estimated from

announced IT capacity (MW) and assumptions about PUE and

usage rates. Although less detailed, this approach allows for

robust projections of future demand.

Based on this framework, Saudi Arabia’s data centers consumed

an estimated 2.8 TWh in 2024, accounting for around 0.85%

of the country’s total electricity consumption in that

year.8 By 2030, electricity consumption by data centers is

projected to increase substantially, with estimates ranging

from approximately 10.16 TWh to 42.23 TWh annually. The

variation reﬂects differences in capacity expansion and the

implementation of efﬁciency practices.

Figure 12 shows that under the baseline growth scenario,

where expansion is limited to general-purpose facilities,

annual consumption could reach 10.16 TWh, or 2.79% of

national demand, reﬂecting steady but manageable growth.

The introduction of AI-driven facilities signiﬁcantly increases

demand. Under the moderate-growth scenario, projected

annual electricity demand could reach 20.15 TWh by 2030.

However, this could be reduced to 17.62 TWh, a 13% reduction,

with advanced cooling and improved design practices. The high-

growth scenario results in the largest increase, with electricity

demand rising to 42.23 TWh annually, equivalent to 11.55%

of projected national electricity demand. This ﬁgure could

be lowered to 36.76 TWh, a 13% reduction, by implementing

efﬁciency standards.

7 PUE: A data center metric that measures energy efﬁciency used by IT equipment.

8 Assuming a PUE of 1.7, a use rate of 0.65, and Saudi Arabia’s total electricity consumption in 2024 as 333 TWh.

AI and Energy: The Future of Data Centers in Saudi Arabia

In all scenarios, electricity demand from data centers will rise

over the decade. The extent of this increase will depend on

the scale of AI-oriented projects and the level of efﬁciency in

facility design and operation. Regardless of the growth scenario

adopting sustainable practices can reduce energy consumption

by 13% compared to conventional approaches.

4.3 Emissions Projections

Electricity consumption is the main driver of data centers’

carbon footprint, with their carbon intensity determined by

the energy mix composition. In Saudi Arabia, where oil and

natural gas are signiﬁcant power sources, this challenge is being

addressed through a commitment to achieve net-zero emissions

by 2060 in line with the Paris Agreement and Saudi Vision

2030 (SGI 2025). The ambition is reinforced by the Saudi &

Middle East Green Initiatives (2025) and an updated Nationally

Determined Contribution that pledges to reduce annual

Figure 12. Historical and projected electricity demand from data centers in Saudi Arabia, 2022-2030.

emissions by 278 Mt CO2 by 2030 (Saudi & Middle East Green

Initiatives 2025). A central pillar of this strategy is the National

Renewable Energy Program, which aims to raise the share of

renewables to 50% of the energy mix by 2030 (Saudi & Middle

East Green Initiatives 2025).

In this study, emissions are estimated under two scenarios:

the conventional fossil fuel mix (41.2% oil and 58.2% natural

gas), and the sustainable national target mix (50% renewables

and 50% natural gas), as shown in Figure 13. For simplicity, we

assume that the capacity mix and the electricity generation

mix are the same. The 50/50 mix aligns with global projections,

where nearly half of the electricity demand from data centers is

expected to be met by renewable sources (Spencer et al. 2025).

With the fossil fuel mix, the electricity used by existing data

centers, in 2024, produces about 1.6 Mt CO2 each year. If growth

follows a traditional and non-AI path, this could rise to 5.81 Mt

CO2 by 2030, while a moderate AI-driven scenario could increase

emissions to 11.48 Mt CO2. In the high-growth scenario, annual

Note: The left axis shows the share of national electricity demand, while the right axis reports absolute consumption in TWh. The shaded bands’ upper limits correspond to

conventional operating practices, while the lower limits reﬂect sustainable operational conditions, detailed in Appendix B.

Source: Authors.

2022

40.0

30.0

20.0

10.0

0.0

10%

12%

2023 2024 2025E 2026E 2027E 2028E 2029E 2030E

Historical Projected

Baseline (non-AI) High-AI growth (band) Moderate AI growth (band)

Year

Electricity consumption (TWh)

Share of national electricity

AI and Energy: The Future of Data Centers in Saudi Arabia

Natural Gas

Share of energy (%)

Oil Renewables

58.2

2022 2030

41.2 50

Figure 13. Saudi Arabia’s energy mix in 2022 compared with the 2030 national target.

emissions could reach 24.02 Mt CO2. However, shifting to that

target energy mix would reduce emissions by roughly 68%

across all scenarios, as Figure 14 shows. For example, moderate

AI growth would emit 3.7 Mt CO2 instead of 11.48 Mt CO2, and

high AI growth would drop from 24.02 Mt CO2 to 7.7 Mt CO2.

These reductions highlight the role of renewable integration in

mitigating the climate impact of digital infrastructure expansion.

In addition to estimating total CO2 emission from data centers, it

is useful to track carbon intensity via carbon usage effectiveness

Figure 14. CO2 emissions from data centers by scenario (Mt).

Source: Authors.

Fossil fuel mix (emission factor [tCO2/MWh] ≈ 0.568)

CO2 emissions (Mt)

2024 (current) Non-AI growth Moderate AI growth High AI growth

1.6

5.81

2.15

11.48

3.7

7.7

24.02

2030 national target mix (emission factor [tCO2/MWh] = 0.21)

AI and Energy: The Future of Data Centers in Saudi Arabia

Legal and Regulatory Environment

Saudi Arabia has developed a comprehensive and coordinated

legal framework for information and communications

technology, data governance, cloud services, sustainability, and

energy efﬁciency, implemented through several institutions. Key

regulations include the Telecommunications and Information

Technology Act (Bureau of Experts of the Council of Ministers

2022), the Personal Data Protection Law (Bureau of Experts

of the Council of Ministers 2021), and the Cloud First Policy

(MCIT 2020), which together regulate data handling, cloud

operations, and data center development. The Data Center

Services Regulations (CST 2023, 2024b) enhance this framework

by requiring licensed operators to prepare sustainability plans

focused on energy efﬁciency, carbon reduction, and electronic

waste management, while promoting environmentally friendly

power and cooling solutions. In addition, the Global AI Hub Per

website in reference Law (Istitlaa 2025) is a major step toward

facilitating cross-border AI deployment through greater legal

certainty and operational ﬂexibility. It deﬁnes three categories

of AI data centers: Private Hubs, Extended Hubs, and Virtual

Hubs, allowing international providers to operate in Saudi

Arabia under regulatory arrangements aligned with their home

jurisdictions.

Strategic Location, Digital Infrastructure,

and Connectivity

Saudi Arabia’s geographic position at the intersection of Asia,

Europe, and Africa gives it strategic advantages for regional

and intercontinental data ﬂow. The Kingdom is connected to

the global internet by 16 active subsea cable systems across

the Red Sea and Arabian Gulf, providing multiple landing points

in ﬁve coastal cities (MCIT 2024). Domestically, ﬁber-optic

infrastructure and 5G networks, covering approximately 65%

of the population in 2024, have improved inland connectivity,

supporting edge computing and other latency-sensitive

applications (MCIT 2024). These infrastructure elements

collectively create a technical foundation capable of supporting

optimized AI data centers with reliable performance.

Energy Advantage and Grid Readiness

Saudi Arabia offers a distinctive energy advantage for AI data

centers, characterized by low electricity prices, an expanding

power grid, and a growing share of renewables. Electricity

tariffs for cloud computing are as low as 18 halalah/kWh (about

$0.048/kWh), while industrial tariffs stand at 20 halalah/kWh

(CUE). CUE measures emissions intensity per MWh delivered

to IT equipment (tCO2/MWh-IT), where a CUE of zero indicates

fully renewable or zero-carbon power. CUE is independent of

load size, allowing for a clear separation between how clean

operations are from how large the load is.

Conventional operations with a PUE of 1.5-1.7 on a fossil fuel

mix yield a CUE of 0.85–0.97 tCO2/MWh-IT, whereas sustainable

operations with a PUE of 1.3-1.5 of the 2030 grid yield a CUE of

0.27-0.32 tCO2/MWh-IT. This is about a threefold reduction in

emissions intensity. CUE highlights that decarbonizing AI data

centers depends primarily on facility efﬁciency (lower PUE) and

the carbon intensity of the power source. High-growth scenarios

raise absolute emissions unless paired with low-CUE operations,

which explains the 68% drop in total CO2 when the cleaner 2030

mix is applied.

4.4 Key Enablers

The growth of Saudi Arabia’s AI data centers is supported

by a strong national strategy, advanced regulation, strategic

geography, and state-backed investment. The energy system

is also very cost-effective, reliable, and has ambitious

sustainability targets. These enablers, together with the

Kingdom’s digital transformation agenda, create an environment

where AI infrastructure can develop competitively and

sustainably. We discuss selected enablers next.

National Strategies

Vision 2030 prioritizes digital transformation and AI as key to

economic diversiﬁcation. Out of 96 Vision objectives, 66 are

directly or indirectly tied to data and AI (SDAIA 2025b). The

National Strategy for Data and Artiﬁcial Intelligence, set by

the SDAIA, aims to position Saudi Arabia among the top 15

countries in AI readiness and to attract about 75 billion SAR

in investments by 2030 (SDAIA 2025a). Complementing this,

the National Cybersecurity Authority (NCA) published the

National Cybersecurity Strategy (NCA 2018) in 2020 to create a

secure and trusted cyberspace for digital infrastructure. Saudi

Arabia also enjoys the world’s lowest levelized costs for both

wind and solar energy, along with targets for 50% renewable

energy and ambitious sustainability goals. There is a clear

complementarity between energy and AI-related policies

that support the country’s competitiveness in hosting and

operating AI data centers.

AI and Energy: The Future of Data Centers in Saudi Arabia

(about $0.053/kWh) (SERA 2025). AI data centers typically fall

within these tariff categories,9 allowing operators to achieve

a total cost of ownership up to 30% lower than in comparable

international markets (Techusiness 2025). The national power

grid is expanding rapidly, with transmission lines expected to

increase from 96,496 km in 2024 to 160,000 km by 2030, and the

number of substations rising from 1,235 to 1,650 (SEC 2025).

These developments are supported by ongoing investment

in battery storage capacity (targeting 48,000 MWh) and

grid modernization projects to enhance thermal efﬁciency

and support future carbon capture systems (SPA 2024).

Furthermore, the Kingdom’s abundant solar irradiance and

emerging wind capacity offer signiﬁcant potential for green

energy integration. Saudi Arabia is actively increasing its

renewable capacity to 130,000 MW by 2030, with 44,000 MW

already tendered and 20,000 MW added in 2023 (KAPSARC

2025). The Kingdom has also set global benchmarks for

renewable energy costs, exempliﬁed by the 2024 Al-Shuaiba

solar project (Techusiness 2025), which achieved one of the

world’s lowest electricity generation costs at $0.0104/kWh

(Bellini 2021). These developments improve the potential

for long-term price stability and decarbonization for energy-

intensive AI infrastructure.

Government Capital and International

Collaboration

The expansion of AI data centers in the Kingdom is driven by

government-backed capital, sovereign investment, and strategic

international partnerships. The Public Investment Fund (PIF)

plays a central role by investing in HUMAIN and other national

AI infrastructure ventures. In addition, strategic collaborations

with global technology leaders, including NVIDIA, AMD, and

AWS, are enhancing computational capability and technology

transfer. These initiatives are supported by the government,

with attractive energy and sustainability offerings, land

allocation, and infrastructure support, designed to attract long-

term investors and speed up deployment.

4.5 Factors that Could

Influence Projections

Electricity demand from AI data centers through to 2030 will

depend on several interacting factors. The most important are

the scale of AI adoption, advances in hardware, improvements

in infrastructure and software, and, over the longer term, new

computing paradigms. These factors will determine whether

energy use rises steeply or can be balanced by efﬁciency gains.

The scale of AI adoption across government, business, and

society will be a major driver of future data center electricity

demand. As AI is used in more critical sectors, the need for

powerful computing will rise. Energy use will also depend

on how complex the tasks are. Highly complex tasks require

more energy due to longer context windows, deeper reasoning

chains, and higher token generation. For example, a simple

chatbot query may consume around 1.55 Wh, while a retrieval-

augmented generation query requires roughly 2.64 Wh, and an

agentic workﬂow requires 8.54 Wh on average (Desroches et al.

2025).

Efﬁciency improvements will play a central role in shaping

the future energy demand of AI data centers. Advances are

happening at several levels:

• Hardware: New generations of processors deliver far more

computing power for each unit of electricity consumed.

Specialized chips designed for AI, like GPUs and TPUs, are

much more efficient than traditional CPU processors. Each

generation also becomes more powerful as capabilities

expand. For example, NVIDIA’s Blackwell B200 GPU offers

better energy efficiency than its predecessors.

• Infrastructure: Data centers are improving electricity

use through better cooling systems and more innovative

design. While much progress has already been made, future

improvements are expected to deliver further gains.

• Software: Smarter algorithms help reduce the amount of

computing needed for complex AI tasks, reducing energy

demand while keeping performance high. For instance,

DeepSeek-R1 reduced energy use with smarter design

9 The exact tariff that data centers will be charged in the Kingdom is yet to be decided. Nonetheless, we refer to these tariffs since they are the only ones that are publicly available.

AI and Energy: The Future of Data Centers in Saudi Arabia

system load or price spikes, operators can reduce peak

consumption without compromising service reliability.

Wider adoption of automated workload scheduling and

real-time pricing mechanisms would smooth demand

curves, lower grid stress, and marginally reduce total annual

electricity use.

Ultimately, while efﬁciency gains can lower the energy required

for a given amount of computation, they also tend to reduce the

cost of computing. This can trigger what is known in the energy

efﬁciency ﬁeld as the rebound effect (or the Jevons Paradox),

where improved efﬁciency lowers costs, thereby stimulating

higher demand and wider adoption of AI applications. As a

result, total energy use may continue to rise even as individual

systems become more efﬁcient.

techniques, like activating only needed parts of the model,

running computations in lower precision, and balancing

work across hardware, resulting in notable efficiency gains.

• New computing paradigms: Looking further ahead,

emerging technologies such as quantum and neuromorphic

computing can offer potential for energy efficiency in

specific AI applications. While their commercial impact is

unknown, these technologies could fundamentally change

the energy landscape of AI.

• Demand response management: Participation of data

centers in demand response programs could significantly

influence the electricity demand of data centers. By

adjusting non-critical computing workloads during high

5.1 Calculating AI Data Center

Project Costs

AI workloads need substantial and ongoing computing capacity,

making data center development both capital intensive

and operationally complex. To evaluate whether expanding

capacity adds value, investors and policymakers need a clear

understanding of the full life cycle economics of building and

operating AI infrastructure. This can be done through a levelized

cost analysis framework, an established method for calculating

costs over a project’s lifespan. The framework identiﬁes the

minimum price at which computing services must operate to

recover capital and operating expenses, thereby assessing

overall project viability. The breakeven cost per unit of compute

is evaluated over the facility’s lifetime, allowing sensitivity

analysis of key factors such as electricity price, PUE, and

load factor.

There is a need for a quantiﬁable economic performance metric

to evaluate data center infrastructure dedicated to AI workloads

This section presents an in-depth cost analysis supporting the assessment of AI data

center demand growth in Saudi Arabia. It examines key cost drivers, including electricity

pricing, energy and computing efﬁciency, and other operational parameters that shape

competitiveness. The analysis outlines the modeling framework applied to estimate

data center costs and then applies it to a detailed case study of Saudi Arabia’s market

conditions.

Cost Analysis of AI

Data Centers in

Saudi Arabia

(Kristiansen Nøland, Hjelmeland, and Korpås 2024). In this

analysis, we estimate the average cost of delivering each unit

of compute over the facility’s operational life. This involves

summing all costs of construction, equipment, electricity, and

maintenance over time and dividing by the total computing

output produced. The results show the minimum price that

must be charged per unit of computing power for the facility

to break even over its lifetime. This follows the logic of the

levelized cost of energy, which compares projects by measuring

lifetime costs per unit of output. Applying it to data centers

provides a consistent means of assessing long-term economic

viability under different operational and market conditions. In

this case, the analysis measures the cost per unit of compute –

typically expressed in dollars per ﬂoating-point operation ($/

FLOP), and often scaled to $/PFLOP or $/EFLOP due to the large

computational output involved.10

Several factors affect the cost of data center projects: PUE, load

factor, computing efﬁciency, CAPEX, operating expenditure

(OPEX), and the weighted average cost of capital (WACC).

Table 5 provides an overview of each factor.

10 A FLOP is a single ﬂoating-point operation, which is used as a measure of computational performance. FLOPs is the total count of ﬂoating-point operations performed to complete a speciﬁc

task or program. The number of FLOPs a system or processor – such as a CPU, GPU, or supercomputer – can perform in one second is used as a measure of computational performance.

PFLOP (petaFLOP) = 1015 FLOPs. EFLOP (exaFLOP) = 1018 FLOPs..

AI and Energy: The Future of Data Centers in Saudi Arabia

5.2 Baseline Results and

Sensitivity Analysis

To evaluate Saudi Arabia’s data center project cost, the analysis

assigns default values to each factor. Table 6 summarizes these

Table 5. Factors that affect the average data center project costs over its lifetime.

Factor Deﬁnition Relevance to data centers Ideal value Best recorded value

PUE The ratio of total energy consumed by

the data center to the energy consumed

by IT equipment only

Measures overall energy

efﬁciency

1, where all the energy

consumption goes to IT

equipment

1.028, by the National

Renewable Energy Laboratory

(NREL) (Van Zandt 2023)

Load factor The ratio of average IT load to peak load

in a data center over a speciﬁc period

Indicates operational efﬁciency

and utilization

100% ~90%

Computing

efﬁciency

The number of PFLOPs/kW AI hardware efﬁciency - ~7.71 PFLOPs/kW by NVIDIA’s

Blackwell Ultra chip (2025)

Electricity price Unit price of electricity Determines ongoing operating

costs

- -

OPEX Annual operating and maintenance

expenses as a percentage of CAPEX

Represents recurring costs

during operation

- -

CAPEX Overnight cost of construction Required initial investment - -

WACC Weighted average cost of capital Reﬂects the cost of ﬁnancing

and investor return expectations

- -

values, along with a justiﬁcation for each selection. It is worth

noting that some values were varied in subsequent sensitivity

analyses to explore policy insights, assess sensitivity to key

factors, and explore trade-offs.

Table 6. Default values used in the case study.

Factor Default value Justiﬁcation

PUE 1.5 Average for existing data centers in Saudi Arabia (range 1.5-1.8)11

Load factor 80% Illustrative mid-range assumption for AI-oriented facilities (70%-90% is typical, though actual use may vary)

(Aterio 2025)

Computing efﬁciency 0.21 PFLOPs/kW Median for AI hardware values (see Table 7)

Electricity price $48/MWh Equivalent to the ofﬁcial tariff for cloud-computing 18 Halalah/kWh (SERA 2025)*

CAPEX $10,000/kW Baseline assumption (Kristiansen Nølan, Hjelmeland, and Korpås 2024), and similar to Equinix planned

hyperscale AI data center (Malik 2025)

WACC 10%

Baseline assumption (Kristiansen Nøland, Hjelmeland, and Korpås 2024). Lower ﬁnancing costs are possible

under domestic conditions

Annual O&M cost12 10% of CAPEX

Project lifetime 15 years

* The exact tariff that data centers will be charged in the Kingdom is yet to be decided. Nonetheless, we use this tariff as an assumption since it is publicly available.

Note: The baseline case applies representative, literature-based parameters for AI data center operations in Saudi Arabia. These values are indicative and intended to illustrate

cost sensitivities rather than to forecast speciﬁc project outcomes. Actual project performance may vary depending on facility scale, technology choice, ﬁnancing conditions, and

operational strategy.

11 S&P Global data.

12 Annual O&M cost: Yearly operations and maintenance cost (OPEX) of total capital expenditures (CAPEX).

AI and Energy: The Future of Data Centers in Saudi Arabia

Figure 15. Annual discounted cost and compute output over time.

Using the default values, the baseline unit cost is estimated at

$0.51/EFLOP, where 1 EFLOP = 1,000 PFLOPs. This represents

the average lifetime cost of computation under typical Saudi

operating conditions, assuming a PUE of 1.5, 80% load factor,

and an electricity price of $48/MWh. Comparable analyses, such

as Kristiansen Nøland, Hjelmeland, and Korpås (2024), report

levelized computing costs of about $1.0-$1.2/EFLOP under

higher electricity prices ($75-$125/MWh), a PUE of 1.12, and a

computing efﬁciency of 0.1 PFLOPs/kW. When normalized to

that same efﬁciency, the equivalent cost in this study would be

approximately $1.07/EFLOP, placing it lower than the median

of the range. This indicates that Saudi Arabia’s AI data centers

could be cost-competitive even with the higher PUE values

typical of warm climates.

To interpret this result, Figure 15 compares the annual discounted

cost with the annual discounted compute output over the project

lifetime. Both measures decrease over time due to discounting,

which reduces the present value of future expenditures and

computing services. The decrease in discounted compute

output does not reﬂect hardware degradation; it represents the

diminishing present value of future computational capacity.

The key result is that most of the project’s economic value is

captured early, when effective usage is high relative to cost. This

is especially true for AI data centers, where rapid technology

cycles and hardware replacement rates make early operational

efﬁciency essential for recovering capital before systems

become outdated.

The unit cost of compute in any given year can be inferred by

comparing the annual discounted cost with the discounted

compute output. Aggregating this relationship across all years

gives us the average lifetime cost of computation. While this

relationship reﬂects standard capital recovery dynamics, it also

highlights why AI data centers exhibit stronger front-loading

effects than most infrastructure assets. Their high capital

intensity, rapid technology turnover, and performance gains in

early operating years mean that utilization achieved in the initial

phase has a disproportionate effect on total cost recovery. In

this sense, early and sustained high load factors are not just

desirable but critical to maintaining cost competitiveness in a

fast-evolving compute market.

Source: Authors.

6,000

5,000

4,000

3,000

Value (discounted)

2,000

1,000

1 2 3 4 5 6 7 8

Year

9 10 11 12 13 14 15

Annual discounted compute (EFLOP) Annual discounted cost ($)

AI and Energy: The Future of Data Centers in Saudi Arabia

Figure 16. Sensitivity to the load factor.

To gain further insights, the correlation between average

data center project costs over its lifetime and multiple factors

is examined. These factors include load factor, computing

efﬁciency, electricity price, and PUE.

Load Factor

As the load factor increases, unit costs decline sharply, as

Figure 16 shows. At low usage, costs exceed $3.40/EFLOP,

but as the load factor rises toward stable operation, they

quickly drop to near $0.5/EFLOP. The curve shows diminishing

marginal cost savings at higher usage levels, meaning that most

efﬁciency gains are achieved when data centers operate near

steady capacity. This ﬂattening of the cost curve highlights the

importance of maintaining consistently high usage to stabilize

both operating costs and power consumption. Operators often

achieve this by smoothing short-term demand ﬂuctuations

through background compute tasks, such as model training,

which help ensure a more constant load on the grid.

Computing Efficiency

Here, we investigate how computing efﬁciency – the

computational output per kilowatt of IT power – affects average

costs. As shown in Figure 17, improvements in computing

efﬁciency signiﬁcantly lower costs. The x-axis ranges between

approximately 0.01 and 0.8 PFLOPs/kW, beyond which the

curve saturates and cost reductions become minimal. At low

efﬁciency, costs are very high, but as efﬁciency increases, costs

fall sharply and eventually ﬂatten. This aligns with industry

trends, as hyperscale operators and AI hardware manufacturers

aim for higher computing efﬁciency to balance throughput

and cost competitiveness. This result is consistent with the

growing interest of hyperscales and AI hardware manufacturers

in achieving greater computing efﬁciency, as shown in Table 8.

In essence, investing in highly efﬁcient compute platforms not

only improves throughput but also enhances long-term cost

competitiveness.

Source: Authors.

3.5

2.5

1.5

0.5

10% 16% 21% 27% 32% 38% 43%

Load factor

49% 54% 60% 66% 71% 77%

Cost ($/EFLOP)

AI and Energy: The Future of Data Centers in Saudi Arabia

Figure 17. Sensitivity to the computing efﬁciency.

Computing efﬁciency

Cost ($/EFLOP)

0.01 0.09 0.17 0.25 0.33 0.41 0.49 0.57 0.65 0.74

Source: Authors.

Table 7. Computing efﬁciency for various AI hardware.

Provider GPU model Year released Computing efﬁciency (PFLOPs/kW)

Intel GPU Flex 140 2022 0.107

GPU Flex 170 2022 0.107

GPU Max 1100 2022 0.048

GPU Max 1550 2023 0.049

AMD Instinct MI300X 2023 0.209

Cerebras (Wang 2024) CS-3 2024 5.43

NVIDIA Tesla T4 2018 0.116

Tesla V100 2017 0.052

DGX A100 2022 0.769

DGX H100 2022 3.137

DGX B200 2024 5.035

GB200 NVL72 2024 6.000

Blackwell Ultra 2025 ~7.71

Electricity Price and PUE: Trade-off Curves

A policy-sensitive lever is the trade-off between electricity price

and PUE. Figure 18 presents the relationship between electricity

price and data center project costs for different PUE scenarios,

ranging from a PUE value of 1.0 (ideal efﬁciency) to 1.7 (less

efﬁcient, reﬂecting more cooling and overhead consumption).

A PUE of 1.1 is typical for AI data centers located in cool

climates, while a PUE of 1.5 represents the average for data

centers in Saudi Arabia.

All scenarios originate from a common baseline at zero

electricity cost, highlighting that ﬁxed costs – CAPEX and

non-electric OPEX – set a minimum threshold for computing

costs. The slope of each line reﬂects sensitivity to electricity

prices, where steeper slopes correspond to higher PUE values.

AI and Energy: The Future of Data Centers in Saudi Arabia

This means that less efﬁcient facilities experience a faster cost

escalation as electricity prices rise. While the data center project

cost is inﬂuenced by power prices, it is less sensitive compared

to factors such as load factor or computing efﬁciency.

The ﬁgure also highlights an important policy and siting insight:

electricity price and PUE interact multiplicatively in determining

compute economics. For example, even at a relatively high PUE

of 1.7, data centers in regions with low industrial tariffs, such

as Saudi Arabia where electricity is priced around $0.048/kWh

(18 halalas/kWh), can still maintain competitive project costs.

This indicates that low-cost electricity markets allow design

trade-offs, allowing developers to operate at higher PUEs within

climatic or infrastructure constraints, while keeping costs at an

acceptable level.

To visualize the combined effects of electricity prices and

PUE on data center projects costs, Figure 19 maps the cost

across different electricity prices and PUEs. At low prices of

$20-$40/MWh, even inefﬁcient data centers with a PUE above

2 can maintain a low cost. For example, at $24/MWh and a PUE

above 2.0, the cost remains below $0.55/EFLOP. This shows that

when electricity prices are not a limiting factor, operators can

tolerate less efﬁcient cooling infrastructure without losing their

competitive edge.

At the Saudi benchmark of $48/MWh for cloud computing,

a facility with a PUE of 1.7 achieves roughly $0.55/EFLOP.

However, if electricity prices double to $96/MWh, the same

facility’s cost would rise above $0.60/EFLOP, representing at

least a 10% increase in costs. This ﬁgure reinforces the idea

that policymakers and developers must consider both factors

together.

In Saudi Arabia, the hot climate makes achieving ultra-low

PUE challenging, but electricity prices are competitive. That

puts the Kingdom in a “competitive zone” on the map: even

with a moderate PUE (e.g., around 1.6-1.8), the costs stay low

Figure 18. Trade-offs between PUE and electricity price.

Source: Authors.

0.7

0.65

0.6

0.55

Cost ($/EFLOP)

0.5

0.45

0.4

0 10 20 31 41 51 61

Electricity price ($/MWh)

71 82 92 102 112 122

PUE = 1.0 PUE = 1.1 PUE = 1.3 PUE = 1.5 PUE = 1.7

AI and Energy: The Future of Data Centers in Saudi Arabia

at current tariffs. Data center designers can choose practical

cooling solutions and still be cost-competitive. As future

infrastructure is planned, maintaining favorable tariffs and

enabling high-efﬁciency design practices will be central to

sustaining this advantage.

5.3 Policy Insights

The cost analysis places Saudi Arabia favorably in the global

landscape for developing AI data centers. The Kingdom’s

electricity tariffs are already among the most competitive in

the world, and are supported by reliable grid infrastructure

and cost stability. This advantage can offset the higher cooling

requirements typical of warm climates and provides a strong

foundation for investments in AI and cloud infrastructure.

Our analysis shows that further reductions in electricity tariffs

would yield only limited beneﬁts, while improvements in energy

efﬁciency and use offer far greater potential to lower overall

compute costs. Enhancing cooling performance, hardware

efﬁciency, and load management is therefore the most effective

path to strengthening competitiveness. Policy could focus on

how energy is used rather than how cheap it is, by promoting

advanced technologies and operational practices that improve

efﬁciency and maintain high, stable usage. From a policy

standpoint, keeping electricity prices stable, establishing clear

performance and efﬁciency standards, and encouraging the

use of the latest in hardware are the most effective levers for

sustaining long-term competitiveness.

Figure 19. Contour map of the cost as a function of PUE and electricity price.

Notes: Each contour line represents a constant cost value, and the shading reﬂects the gradient of cost.

The plot includes several vertical reference lines to guide interpretation: one at $24/MWh (half of the cloud computing tariff), one at $48/MWh (the benchmark tariff for cloud

computing in Saudi Arabia), and one at $96/MWh (double tariff).

1.4

1.5

1.6

1.7

1.8

2.0

1.9

Power usage effectiveness (PUE)

2.1

20 40 60

Electricity price ($/MWh)

80 100 120

0.40

0.44

0.48

0.52

0.56

0.60

0.64

0.68

0.72

0.76

0.59

0.50

0.46

0.41

6.1 Challenges

Regulatory and Policy Pressures

Governments and regulators increasingly impose new rules

on data centers to meet environmental objectives and

infrastructure planning needs. In mature markets, there are

limits on new builds and binding sustainability requirements,

such as disclosing annual energy and water use, stricter

efﬁciency standards, and commitments to 100% renewable

power. These policies reﬂect a broader regulatory trend in which

data centers must align with national climate targets or face

permitting hurdles and growth caps. For operators, this raises

the bar on compliance and increases costs. Meeting new rules

will require larger investments in renewable procurement,

higher-efﬁciency cooling, heat-recovery systems, and better

measurement and reporting.

Some countries’ data protection and localization regulations

require speciﬁc data to be stored domestically. At the same

time, renewable-energy compliance mechanisms, such as

RECs and green-power mandates, are becoming increasingly

The expansion of AI data centers has the potential to create economic and technological

opportunities. However, as AI data centers proliferate and increase their capacities, vari-

ous risks and challenges must be considered. Challenges reﬂect structural and technical

difﬁculties to AI data center growth, while risks highlight potential adverse outcomes. This

section analyzes both dimensions globally, not tied to any speciﬁc region.

Global Overview of AI

Data Center Challenges

and Risks

common, requiring data center operators to buy or generate

certiﬁed renewable electricity to meet sustainability targets.

Overall, regulatory pressures are rising on multiple fronts

– spanning energy sourcing, water use, carbon emissions,

and data governance – creating a more complex compliance

landscape for developers. Failure to meet these evolving

standards may raise costs or slow expansion, particularly in

markets where environmental and energy regulations are

tightening.

Land Use and Infrastructure Bottlenecks

As AI data centers increase, physical infrastructure and land

availability become key constraints. These facilities need more

than just open space; they require particular siting conditions

with access to high-capacity power grids, ﬁber-optic networks,

cooling resources, and stable geology. Globally, this has led

to clustering in major metropolitan hubs, often causing local

bottlenecks in grid capacity, land zoning, and ﬁber connectivity.

While land is not inherently scarce in many countries, the

challenge lies in ﬁnding sites that meet technical criteria for

reliability and latency, and infrastructure requirements like

AI and Energy: The Future of Data Centers in Saudi Arabia

electric vehicles (Spencer and Singh 2024). The International

Energy Agency (IEA) notes that delays to grid- connections are

a major risk, caused by lengthy permitting processes and long

transmission lead times. Connection queues in the U.S. average

one to three years, and in some areas, such as Northern Virginia,

can exceed seven years (Spencer et al. 2025). McKinsey likewise

highlights “time to power” as the main concern for data center

operators when building new sites (Green et al. 2024).

Beyond grid permitting, bottlenecks in equipment and

power generation are becoming critical barriers. Transformer

manufacturing lead times have more than doubled since 2020,

with utilities and developers now facing wait times of more

than two years for high-voltage units (IEA 2024). Similarly, gas

turbine supply chains are facing unprecedented backlogs, as

manufacturers prioritize delivery for grid-stability and industrial

projects, further delaying the deployment of on-site backup

power. These constraints exacerbate delays for connection

and make it increasingly difﬁcult to synchronize grid access,

equipment delivery, and construction timelines.

Together, these challenges could delay projects, increase

costs, or force reliance on carbon-intensive backup generation,

undermining both growth and sustainability targets. Securing

an adequate supply of grid infrastructure and critical

components, along with efﬁcient permitting processes, is now a

strategic necessity for sustaining data center expansion in

any region.

Environmental Footprint

The growth of data centers raises concerns about environmental

sustainability. Today, data centers account for nearly 0.5% of

global fuel-combustion CO2 (about 180 Mt) and are projected

by the IEA to reach about 1% by 2030, the equivalent of about

3% of global electricity (Spencer et al. 2025). Much of this

environmental impact comes from the energy demand of

servers and cooling systems, particularly if that energy is

generated from fossil fuels.

Water consumption is also a sustainability concern due to the

large water footprint of data centers. Most hyperscale facilities

rely on water-based cooling to dissipate heat generated by

servers. A small 1 MW data center can consume as much as

25.5 million liters of water per year (Mytton 2021), equivalent

to the yearly use of more than 60 U.S. households (Sharma

substation access, road transport, and chilled-water cooling

systems. This challenge is compounded when backup power

systems and cooling plants are also required.

Moreover, as data centers integrate renewable energy sources,

the land required for on-site generation – such as solar PV –

can become a signiﬁcant constraint, particularly near urban

or industrial areas. In those cases, relying on grid connections

to large-scale renewable plants is often more practical and

land-efﬁcient than developing generation on site. This allows

renewable power to be sourced from utility-scale plants in

high-resource areas while minimizing land-use footprint and

logistical complexity at the data center itself.

Talent Shortages and Workforce Constraints

The AI data center sector faces a global shortage of skilled

professionals in areas such as electrical engineering,

cybersecurity, network design, and AI infrastructure. The

U.S. alone faces a projected shortfall of more than 300,000

workers by 2025 (Levine 2025). Similar gaps exist in Europe

and Asia, where training opportunities have not kept pace with

hyperscale development.

While Saudi Arabia invests in digital skilling and localization

through Vision 2030, a signiﬁcant talent gap remains.

Specialized training for AI workloads and advanced cooling

systems (e.g., liquid cooling) is not yet widespread. Relying on

foreign labor may be unavoidable in the short term, and long-

term growth depends on developing a skilled local workforce.

Expanding university programs, establishing vocational centers

focused on data center operations, and fostering industry-

academic collaboration will be essential to securing operational

continuity and supporting domestic employment goals.

6.2 Risks

Energy Supply Constraints

The rise in the number of AI data centers is placing signiﬁcant

strain on electricity grids and raising concerns about whether

energy supply and grid infrastructure can keep up. Large

hyperscale campuses can demand 100 MW or more – equivalent

to the power use of a steel mill or hundreds of thousands of

AI and Energy: The Future of Data Centers in Saudi Arabia

2024). Larger U.S. data centers can draw as much water as 2

million households annually (Sharma 2024). Such demands

are especially problematic in dry regions. With Saudi Arabia’s

desert climate, it is essential to balance large-scale data center

growth and cooling needs with sustainable water management

practices.

Investment Uncertainty

Massive capital investment is pouring into data center

infrastructure to meet the growing demand for AI and cloud

computing. However, the sector’s long-term returns are

not guaranteed. This AI-driven expansion has been likened

to a “gold rush” (Graham, Rutten, and Yashkova 2024),

with companies racing to build new facilities based on the

assumption of sustained exponential demand growth. The

risk is that projected demand is overestimated. If the economy

slows down or AI fails to deliver the expected business value,

hyperscale providers could reduce spending, which would slow

the pace of expansion.

Another concern is the possibility of oversupply. With so many

new players and investors entering the market, some regions

may have more capacity than needed, which could push

down occupancy rates and prices. At the same time, rising

construction costs and high interest rates add ﬁnancial pressure

to new projects. Together, these factors mean that while growth

prospects appear to be strong, the long-term proﬁtability and

usage rates of new data centers are uncertain.

Geopolitical and Supply Chain Risk

The global availability of high-performance chips, particularly

GPUs and TPUs, is a critical bottleneck for scaling AI data

centers. AI infrastructure relies on a few global chokepoints:

semiconductor fabrication primarily in Taiwan, South Korea,

and the U.S., and access to rare earth elements and batteries

sourced mainly from China and Central Africa. This makes

the sector highly vulnerable to geopolitical tensions, export

controls, and supply chain disruptions, all of which could

severely impact the availability of GPUs, memory, and power

backup systems (Spencer et al. 2025).

Recent U.S. export bans on high-performance chips to China

and other nations show how quickly access can be constrained.

Globally, lead times for data center hardware have already

increased, with critical components like power transformers

facing delays of 18-24 months (Green et al. 2024). To

mitigate this vulnerability, countries may develop diversiﬁed

procurement strategies, localized assembly of non-sensitive

hardware, and long-term contracts with suppliers. Building

regional stockpiles of critical components or investing in MENA-

wide programs of supply chain resilience could strengthen the

Kingdom’s position as a secure and reliable infrastructure host.

7.1 Techniques for Enhancing

Operational Efficiency

Facility Design and Modular Architecture

Modern AI data centers are moving from single mega-buildings

to prefabricated “blocks” that bolt together like Lego. Each

block ships with power distribution, a liquid-ready cooling loop,

and sensors, so operators can add extra capacity when demand

grows. Factory-set airﬂow paths keep hot and cold air apart and

remove the “dead zones” that waste energy in custom halls. As

This section discusses different ways of reducing carbon emissions and enhancing the

operational efﬁciency of AI data centers. We also provide an analysis of selected existing

policies globally targeting more sustainable and efﬁcient deployment and use of AI data

centers.

Toward Sustainable

and More Efﬁcient AI

Data Centers

a result, new modular server halls are already recording PUE

values of approximately 1.2 (EPRI 2024), about 30% better than

the global average of 1.55 in 2022 (Duncan et al. 2024).

Using data center infrastructure management tools that

monitor every rack in real time and shift workloads or fan

speeds can help to keep each one running at its optimal

performance. Half the facility managers surveyed expect this

alone to increase efﬁciency within ﬁve years (Donnellan 2023).

When combined with using virtual servers, which has reduced

physical server counts by 30%-40% in pilot projects, idle energy

can fall to a single-digit share of total IT load (EPRI 2024).

AI and Energy: The Future of Data Centers in Saudi Arabia

Advanced Cooling Techniques

Traditional air cooling is becoming insufﬁcient for high-density

AI computing, particularly with GPUs exceeding 1 kW of power

per chip (Spencer et al. 2025). A conventional air cooling

system can account for up to 40% of a data center’s electricity

use (Duncan et al. 2024). Liquid cooling technologies, such as

direct-to-chip and immersion systems, offer substantial thermal

and energy efﬁciency improvements, reducing power use by

more than 50% while also enabling higher compute density

and less ﬂoor space usage (Duncan et al. 2024; EPRI 2024).

These improvements can lower PUE to below 1.1 (EPRI 2024).

However, as liquid systems often need signiﬁcant volumes of

water, they bring sustainability challenges. To mitigate those

challenges, rainwater harvesting and geothermal cooling

have been proposed as alternative solutions that could reduce

both energy and water demand for AI data center cooling

(Stansbury et al. 2025).

AI Optimized Hardware: ASICs and

Accelerators

The energy efﬁciency of hardware can be improved through

specialized AI accelerators, including application-speciﬁc

integrated circuits (ASICs) and TPUs. These chips are designed

for AI tasks that can be carried out in parallel with each other,

such as matrix multiplications, which dominate deep learning

workloads. Epoch AI ﬁnds that the energy efﬁciency of frontier

machine learning chips has been rising by approximately

a factor of two every two years (Rahman 2024). Together,

such accelerators form the basis of sustainable AI hardware

roadmaps in hyperscale projects, cutting kWh per computation

even as model sizes grow.

Example: Meta’s Data Center Heat Recycling in Denmark

Meta’s aim is to run efﬁcient, certiﬁed data centers on 100% clean and renewable energy. As of 2024, Meta’s contracted

portfolio is more than 11,700 MW of renewables (Meta 2024) and its hyperscale data center in Odense, Denmark, has become

a showcase of circular energy innovation. The data center is linked to the city’s district heating grid so that warm air from

the server halls is routed to a dedicated heat-recovery system. The resulting surplus heat is delivered to the district heating

network and distributed to homes and businesses through Odense’s existing pipes.

According to Meta, 165,000 MWh per year of surplus heat from Odense’s server halls is now supplied into the local district

heating system, reaching up to 9,000 households (Meta 2025). In addition, the campus achieved LEED® Gold and was named

Green Data Centre of the Year in 2021, reﬂecting an efﬁciency-ﬁrst design that enables reliable heat capture.

AI and Energy: The Future of Data Centers in Saudi Arabia

AI Workload Efficiency: Model Prospective

Hardware improvements are increasingly complemented by

algorithmic and architectural techniques that reduce energy per

training run or inference pass. These include (EPRI 2024):

• Model pruning: A technique to remove unnecessary or

redundant weights, neurons, or even entire attention heads

in a trained neural network, leading to multi-fold reductions

in computing and electricity per iteration.

• Quantization: A method to convert model weights to

lower precision formats, such as 8-bit integers. It helps

to accelerate inference and reduce memory and power

consumption on embedded devices.

• Knowledge distillation: An approach to develop a smaller,

more manageable model that reflects the functionalities of

a larger one, reducing computational requirements.

Example: Microsoft’s Waterless Cooling Innovation

As AI workloads surged, Microsoft faced a growing sustainability risk: traditional cooling consumes large volumes of potable

water, which is especially problematic in arid regions with high community and regulatory expectations (Solomon 2024).

In August 2024, Microsoft introduced a “zero-water” cooling architecture for AI-optimized centers, replacing a traditional

evaporative system with chip-level direct liquid cooling in a closed loop. Once ﬁlled during construction, the coolant circulates

continuously between servers and heat exchangers to maintain precise temperatures without using additional water.

Additionally, Microsoft pairs the loop with high-efﬁciency chillers to minimize the energy trade-off of replacing traditional

evaporation and operates at higher server inlet temperatures.

Each method is projected to save over 125 million liters of water per year compared with traditional ones, protecting

local water sources and reducing dependence on municipal supplies (Solomon 2024). Microsoft has steadily improved

the effectiveness of water usage in its data centers to 0.30 L/kWh, a 39% improvement since 2021 and an 80% reduction

compared with its ﬁrst-generation data centers in the 2000s. The method maintains reliable temperatures for dense AI loads

while keeping power use minimal, enabling expansion into arid areas, de-risking permitting, and strengthening Microsoft’s

ability to scale AI sustainably.

These methods help cut compute time, GPU hours, and

emissions without a signiﬁcant loss in performance.

AI Workload Efficiency: Task Scheduling

Running some computing jobs at different times, or in different

data center locations, can reduce the strain on the power grid.

Deloitte’s analysis shows that if AI data centers trim just 1% of

their electricity use during the busiest hours, grid operators

could add about 126,000 MW of new data center load without

major network upgrades (Stansbury et al. 2025). According to

Deloitte’s survey, 68% of industry executives accept this small

cut in exchange for faster grid connections (Stansbury et al.

2025). By integrating workload schedulers that respond to

electricity-price or congestion signals in real time, operators can

avoid costly capacity expansions, lower their carbon footprint,

and even earn demand-response revenues, making task mobility

a scalable option for improving sustainability.

AI and Energy: The Future of Data Centers in Saudi Arabia

7.2 Options for Decarbonization

Long-Term Renewable Power Agreements

Data center operators sign long-term power-purchase

agreements (PPAs) with new wind and solar electricity

generators, giving then a steady ﬂow of green electricity at

a ﬁxed price for a speciﬁed period and helping new projects

secure ﬁnancing. Corporate PPAs support capacity of around

120,000 MW of operational renewables globally, more than

30% of which is accounted for by companies operating data

centers. In fact, those PPAs cover over 20% of the estimated

415 TWh global data center electricity demand in 2024. An

additional 60,000 MW of PPA-related capacity projects are being

developed, much of which is already sold to the same sector

(Spencer et al. 2025).

24/7 Carbon-Free Energy Matching

Traditional approaches often rely on annual clean energy credits

to balance total yearly consumption. Changing to hourly carbon-

free matching would need ﬂexible solutions, like batteries that

can store and shift solar or wind power, steady clean sources

(e.g., nuclear or geothermal that run continuously), and, in

some cases, fossil-fuel plants equipped with carbon capture

and storage (EPRI 2024; Diamant 2022). A real-world pilot

by Google’s Belgian campus illustrates the mix, combining

rooftop PV panels with a 2.75 MW/5.5 MWh lithium-ion battery,

which provides a smooth onsite supply and supports grid

frequency services (Spencer et al. 2025). Together, hourly

tracking, batteries, and steady clean plants allow operators to

match every kilowatt per hour consumed with a carbon-free

supply source.

Small Modular Reactors (SMRs)

SMRs have a power capacity of up to 300 MW per unit, about

one-third of a traditional nuclear power reactors’ capacity (Liou

2023). Announced projects linked to data center supply would

add up to 25,000 MW worldwide (Spencer et al. 2025), mostly

in the U.S. Vendors such as NuScale have designs that come in

250-600 MW blocks, small enough to power a single hyperscale

campus (EPRI 2024).

Fuel-Cell Backup Systems

These systems use hydrogen fuel cells, converting the chemical

energy in hydrogen to electricity with few byproducts, such

as water and heat (EERE 2014). For example, American

Electric Power has agreed to purchase up to 1,000 MW of

Bloom Energy’s solid-oxide stacks for data center customers,

which will replace diesel generators during power outages

(Spencer et al. 2025).

Advanced Geothermal

By drilling deep into hot, dry rock, heat is transferred from

underground by ﬂuid running through sealed pipes in the

drill well to a turbine on the surface. Because the well is a

closed circuit, it can be sited almost anywhere, making it an

attractive carbon-free power option for data center campuses

(Spencer et al. 2025).

Example: Google’s Geothermal Energy for Continuous Clean Energy

Google aims to power its data centers on 24/7 carbon-free energy by 2030 (Google 2025). As intermittent wind and solar

alone could not deliver that round-the-clock coverage, Google partnered with Fervo Energy in 2021 on the world’s ﬁrst

corporate agreement to develop an enhanced geothermal project. In November 2023, the plant began an enhanced

geothermal pilot, with carbon-free electricity ﬂowing to the local grid serving Google’s Nevada data centers. Fervo’s project

test showed commercial viability, recording a 30-day ﬂow that enabled 3.5 MW of electric production, achieved via horizontal

well pairs, 191°C reservoir temperatures, and real-time ﬁber-optic monitoring. Building on this success, Google expanded the

partnership in 2024 by contracting 115 MW through NV Energy’s Clean Transition Tariff – a move which, after full commercial

deployment, will enhance geothermal generation by nearly 25 times that of the pilot.

AI and Energy: The Future of Data Centers in Saudi Arabia

February 2024, U.S. lawmakers introduced the AI Environmental

Impacts Act, which directs the Environmental Protection Agency

and the National Institute of Standards and Technology to

assess the environmental impact of AI in collaboration with

academic, industry, and civil society stakeholders. The Act also

includes a voluntary reporting framework for AI developers

and operators as a ﬁrst step toward common measurement

standards (Crawford 2024). Although there is no binding federal

efﬁciency mandate, the U.S. leads the Net-Zero Government

Initiative to cut government-operational emissions to net-zero

by 2050 (Ofﬁce of the Federal Chief Sustainability Ofﬁcer 2025).

Complementing this, America’s AI Action Plan, published in

July 2025, links AI competitiveness to infrastructure scale,

energy, and regulation. It proposes faster permitting for data

centers under the National Environmental Policy Act (NEPA),

7.3 Selected Regional Policies

As AI data centers proliferate globally, governments and

institutions are responding with regulations focused on the

environment and sustainability to align AI data centers with

climate goals using performance mandates (such as PUE

targets), zoning and reporting requirements, and incentive

schemes. Table 8 highlights key policies and frameworks related

to AI data centers in the U.S., United Kingdom, Singapore, China,

United Arab Emirates, and other countries.

United States

U.S. policy on data centers is evolving through a combination

of federal initiatives, state measures, and strategic planning. In

Figure 20. Sustainable AI data center policy dimensions.

Energy efﬁciency standards

Control of energy performance through PUE thresholds and

infrastructure benchmarks.

Renewable energy and carbon limits

Mandatory or voluntary transition to adopt low-carbon operations

and meet emissions goals.

Infrastructure siting and grid resilience

Location-based incentives or constraints to reduce stress on grids

and urban zones.

Energy reuse and heat recovery

Encouraging waste heat recovery and reuse or integration with

district heating.

Transparency, disclosure, and reporting

Mandatory disclosure of energy usage, emissions, and sustainability

performance.

Development controls and growth caps

Temporary bans, moratoriums, or permit limits on expansion in

speciﬁc areas.

Water and environmental resource management

Response to drought and water scarcity, and site-speciﬁc

sustainability.

AI and Energy: The Future of Data Centers in Saudi Arabia

more federal land for green data centers, modernizing the

national grid for AI-scale loads, and prioritizing energy sources

like nuclear and geothermal to power next-generation AI

infrastructure (White House 2025).

Policies in individual states vary. Some offer tax incentives for

data centers that meet high energy-efﬁciency or renewable-

procurement standards, while others impose limits on backup-

generator emissions or power usage in constrained grids. For

example, since 2014 California has required data centers in the

state that are larger than 1,000 sq ft to report PUE annually, and

any with a PUE above 1.5 must reduce it by at least 10% per year

until reaching 1.5 or lower (California Department of General

Services 2014). Texas Senate Bill 6 sets strict rules for large

electricity users, including data centers. It requires transparent

load planning, on-site backup power, and cost-sharing for

upgrades to the grid. It also authorizes the Electric Reliability

Council of Texas to manage loads during grid emergencies and

to use remote shutoff capabilities. The goal is to protect grid

reliability for residential customers while still encouraging

business growth (Paz 2025).

United Kingdom

The United Kingdom is steadily tightening sustainability

requirements for data centers in line with its target of net-

zero by 2050, set in the UK Climate Change Act. A key policy

mechanism is the Climate Change Agreement (Environment

Agency 2022) for the data center sector, a voluntary scheme

that grants energy tax relief in exchange for meeting efﬁciency

targets, with penalties for non-compliance. Large data centers

also fall under the UK Emissions Trading Scheme (Department

for Energy Security & Net Zero 2025), which sets prices on

emissions and creates ﬁnancial incentives to reduce carbon

intensity.

The Streamlined Energy and Carbon Reporting framework

(Department for Education and Education and Skills Funding

Agency 2025) mandates operators to publicly disclose their

energy usage and carbon emissions, encouraging the adoption

of sustainable practices such as integrating on-site and off-site

renewable energy, efﬁcient cooling technologies, and low-

carbon power procurement.

The European Union

The EU has advanced regulatory efforts to align data centers with

its targets for climate-neutrality and high energy-efﬁciency by

2030. The 2019 EU regulation on eco-design standards for servers

and data storage imposes standards for energy efﬁciency and

sustainable materials, through improved design for durability,

repairability, and recyclability (European Commission 2019).

The Corporate Sustainability Reporting Directive, published in

2022, requires large and publicly listed companies to report

on greenhouse gas emissions, including those associated with

IT infrastructure and third-party data providers (European

Parliament and Council of the European Union 2022). The

Energy Efﬁciency Directive, revised in 2023, requires all

data centers over 500 kW to report metrics on operational

sustainability, including energy use, PUE, water consumption,

renewable share of energy, and waste-heat reuse (European

Parliament and Council of the European Union 2023). It also

requires operators to implement certiﬁed energy management

systems such as ISO 50001.

A new Disclosures Delegated Act in 2024 deﬁnes a standard

energy rating scheme and transparency rules for publication

(European Commission 2024). Complementary EU efforts

include a voluntary Code of Conduct for Data Centre Energy

Efﬁciency (Acton, Booth, and Paci 2025) that promotes best

practices across the industry and include sustainable ﬁnance

rules (European Development Finance Institutions 2024) (EU

Taxonomy) that direct investments toward eco-friendly projects

and initiatives.

China

China has launched major initiatives to green its fast-growing

digital infrastructure. The government’s Eastern Data and

Western Computing initiative, proposed in 2022, aims to

relocate data centers to the western region, and to take

advantage of natural cooling, clean energy, and cost-effective

resources (Zhang et al. 2024). It includes building 10 high-

density, energy-efﬁcient, low-carbon data center clusters in

eight hubs (Mengzhuo and Zhewen 2024). This is expected to

reduce emissions from the data center sector by 16%-20% by

2030 (Zhang et al. 2024).

China also released the National Action Plan for New Computing

Infrastructure in October 2023, prioritizing the construction

of “green, low-carbon” computing systems (Chinese Ministry

of Industry and Information Technology et al. 2023). The plan

includes the adoption of advanced cooling technologies, such as

AI and Energy: The Future of Data Centers in Saudi Arabia

In 2024, the Green Mark for Data Centres certiﬁcation was

enhanced to include broader sustainability metrics such

as energy efﬁciency in IT operations, intelligent systems

integration, carbon reduction strategies, and renewable

energy readiness (Building and Construction Authority

2024). Additionally, and EMA launched the Green Data Centre

Roadmap, which outlines Singapore’s long-term ambition to

lead in AI-ready, tropical climate-optimized, low-carbon data

centers, including innovations like carbon-intelligent scheduling

and modular retroﬁts (Hui Tian 2024).

United Arab Emirates

The UAE’s Moro Hub is a regional benchmark for sustainable

data center development. This 100 MW campus was launched

in 2022 by Digital DEWA and is powered entirely by the

Mohammed bin Rashid Al Maktoum Solar Park, earning a

Guinness World Record as the world’s largest solar-powered

data center (Intel 2025). This was achieved with the help of

underlying federal and local sustainability standards, such as

the Dubai Green Building Regulations (Goverment of Dubai

2023), issued in 2010 and replaced in 2020 by the Al Sa’fat Green

Building Rating System, which set minimum standards for green

design and energy efﬁciency for new buildings, including digital

infrastructure.

Abu Dhabi developed the Estidama Pearl Rating System in

2010, a framework to promote sustainable urban development

and building practices, focusing on environmental, economic,

cultural, and social aspects (Abu Dhabi Urban Planning Council

2016). It consists of ﬁve certiﬁcation levels and eight criteria

categories, including energy efﬁciency and renewable energy

sources.

Together, these policies support the UAE’s Net-Zero 2050

pathway for data centers.

Other Countries

Several countries are adopting targeted regulations to manage

data centers’ energy, environmental, and spatial impacts.

Germany’s Energy Efﬁciency Act requires all new data centers

to have a PUE of 1.2 by 2026, while existing facilities must

meet a PUE of 1.5 by 2027 and improve it further to 1.3 by

2030, alongside a complete transition to renewable energy

by 2027. Germany also introduced the Energy Reuse Factor,

requiring operators to reuse at least 10% of energy (waste

heat) by 2026, rising to 15% by 2028 (Spencer et al. 2025).

Table 8 summarizes different countries’ PUE targets.

natural and liquid cooling, and improved computing efﬁciency,

which, together, support China’s two carbon goals: to peak

carbon emissions by 2030 and to reach carbon neutrality by

2060 (Xu et al. 2023). In mid-2024, China set speciﬁc targets for

data centers: reduce the average PUE to below 1.5 by 2025 and

raise the renewable energy usage rate by 10% annually (Xinhua

2024). These targets are enforced through energy efﬁciency

standards and upgrades to existing facilities.

Australia

Australia has taken signiﬁcant steps toward advancing

sustainable data center governance. The National Australian

Built Environment Rating System (NABERS), ﬁrst introduced

in 1998, is the world’s only mandatory energy performance

labeling scheme for buildings, including data centers. Under

NABERS, data centers are rated from 1 to 6 stars, with a PUE

scale ranging from 2.42 (1- star) to 1.07 (6 stars) (Spencer et al.

2025). The Net-Zero Government Operations Strategy requires

that by July 2025, all federal government workloads must

operate in data centers rated at least 5-star, corresponding

to a PUE of approximately 1.4. This is part of Australia’s goal

to achieve 100% renewable electricity across government

operations by 2030 (Ghadially 2025).

In 2025, the Australian Energy Market Commission strengthened

the country’s grid resilience further by approving new rules to

streamline renewable energy connections and set performance

obligations for large energy users, particularly data centers

and hydrogen projects. These rules would require data centers

to remain stable during grid disturbances and support system

security, reﬂecting their growing share of national electricity

demand (Ghadially 2025).

Singapore

Due to land scarcity and energy constraints, Singapore has

adopted a strategic, sustainability-ﬁrst approach to data

center growth. In 2019, it paused permits for new data center

developments. Permits were restarted in 2022 through a pilot

framework that favored proposals demonstrating exceptional

energy efﬁciency and sustainability performance (Spencer et al.

2025). New data centers must meet a maximum PUE of 1.3 and

obtain the BCA-IMDA Green Mark Platinum certiﬁcation for data

centers, which is the highest tier in Singapore’s green building

rating system (Yeo 2022). Singapore’s national Green Data

Centre Standard was updated in 2020 to align with ISO 50001 to

standardize energy-management practices (Spencer et al. 2025).

AI and Energy: The Future of Data Centers in Saudi Arabia

21% of national electricity consumption by 2020, a freeze was

enacted in the Dublin area until 2028 (Duncan et al. 2024). The

Netherlands enacted moratoriums in 2019 to reevaluate its data

center strategies due to rapid growth in digital infrastructure.

These restrictions were later lifted under revised conditions that

imposed tighter standards for energy efﬁciency and land use.

Environmental factors such as water use also shape policy. In

Chile, authorities partially reversed Google’s permit to build

a data center due to local concerns over water scarcity– a key

issue in a region already stressed by drought. A similar debate

is happening in Uruguay, where community and environmental

groups have raised alarms over the water demands of planned

hyperscale infrastructure.

South Korea takes a different approach by using location

incentives to steer data center development away from areas

with grid congestion. The government offers a 50% reduction

in electricity costs for data centers built outside the heavily

populated Seoul metropolitan area, aiming to ease grid pressure

and support more balanced regional developments

(Spencer et al. 2025).

South Africa’s National Data and Cloud Policy designates zones

for future data centers, ensuring that expansion aligns with grid

capacity and national infrastructure planning goals.

Beyond location-based incentives, several countries are imposing

moratoriums and stricter permitting policies to manage the

growth of data centers. In Ireland, as data centers reached

Table 8. PUE mandates for selected countries.

Region PUE (2023) PUE mandate

U.S. ~1.4 (avg) -

Australia 1.44 1.4 by 2025

China 1.56 1.5 by 2025

France 1.36 40% building energy cut by 2030

Germany 1.42 1.2 (new, 2026), 1.3 (all, 2030)

Japan 1.53 <= 1.4 since 2022

California (U.S.) 1.21 <= 1.5 since 2014

Sources: IEA (2025), Spencer et al. (2025), and Masanet et al. (2024).

Conclusion and

Recommendations

The rapid rise of AI is reshaping the global digital landscape, with data centers at the

heart of this transformation. AI-driven workloads demand powerful and robust computing

systems, reliable energy, and sustainable design, creating both opportunities and

challenges for countries seeking to build leadership in this new era.

This trend presents a unique chance for Saudi Arabia to position

itself as a global hub for AI-ready infrastructure. The Kingdom

beneﬁts from a clear national vision, low-cost energy, and

strong government support. Large-scale projects already

underway, such as HUMAIN’s AI factories and NEOM’s Oxagon

campus, show how quickly the sector is advancing. If current

plans are realized, Saudi Arabia could capture a signiﬁcant share

of global AI computing capacity by the early 2030s, directly

supporting the ambitions to drive digital transformation and

diversify the economy.

However, this study also shows that the path ahead is not

without risks. AI data centers place heavy demands on

electricity systems and water resources. Under a high-growth

scenario, data center electricity demand could increase more

than tenfold by 2030, representing up to 11% of national

electricity demand. Without sustainable practices, this

expansion could further increase emissions and strain

natural resources. By adopting advanced cooling systems

and integrating renewable energy, electricity use could be

cut by almost 13% and carbon emissions reduced by

68%, thus achieving growth without sacriﬁcing

environmental goals.

The economic outlook is also shaped by cost competitiveness.

Thanks to affordable electricity and a favorable location, Saudi

Arabia holds an advantage in the cost analysis. Yet, to sustain

this advantage, the Kingdom must continue to invest in energy

efﬁciency and workforce development, and must keep its

competitive tariffs. Global risks, such as shortages of advanced

chips or geopolitical tensions, also underline the need for

careful planning and diversiﬁed partnerships.

Overall, we ﬁnd that Saudi Arabia stands at a crossroads. By

combining its energy strengths with sustainable practices

and continued investment in innovation, the Kingdom can

build a resilient AI data center ecosystem that both supports

its economic ambitions and contributes to the global digital

progress. Success will depend on balancing three priorities:

ensuring a reliable and clean energy supply, building world-

class digital infrastructure, and embedding sustainability at

every stage of development.

To ensure global competitiveness and sustainable expansion

of AI data centers in Saudi Arabia, as well as a balance

between the three priorities above, several actions need to be

considered across four dimensions: investment, innovation and

development, operations, and governance and policy.

AI and Energy: The Future of Data Centers in Saudi Arabia

Investment

• Invest in efficient technologies: Support hardware such

as next-generation GPUs and efficient servers that deliver

more computing power per unit of energy.

• Create AI-ready investment zones: Prepare dedicated sites

complete with reliable grid connections as a package with

fiscal incentives for investors who commit to sustainability

and renewable sourcing of energy.

Innovation and Development

• Support local R&D: Establish specialized national research

centers dedicated to AI data centers, enabling local

development of advanced cooling, efficiency solutions,

and sustainable infrastructure tailored to Saudi Arabia’s

needs.

• Drive innovation partnerships: Establish partnerships

between universities, research centers, and international

firms to create new solutions for efficient AI systems and

infrastructure.

Operations

• Prioritize early utilization: Maximize utilization in new

data centers in the early stages, as their economic value is

highest in the first years of deployment and operation.

• Maximize efficient use: Require operators to maintain at

least high utilization by implementing workload scheduling

strategies on flexible tasks such as AI training during off-

peak hours to minimize idle infrastructure.

• Adopt sustainable practices: Encourage operators to

recycle heat, reuse water, and adopt circular resource

strategies where possible.

Governance and Policy

• Preserve competitiveness with standards: Maintain

low electricity tariffs, but link incentives and permits to

efficiency benchmarks and commitments to renewables.

• Set efficiency targets: Adopt realistic benchmarks (e.g., PUE

≤1.5 by 2030) tailored to Saudi Arabia’s needs and conditions.

• Align with the energy transition: Coordinate data center

expansion with national grid investments in solar, wind, storage,

and transmission to secure a clean, reliable supply of energy.

References

Abu Dhabi Urban Planning Council. 2016.

The Pearl Rating

System for Estidama: Public Realm Rating System: Design &

Construction, Version 1.0

. December. Abu Dhabi Urban Planning

Council. https://www.dmt.gov.ae/-/media/Project/DMT/DMT/E-

Library/0001-Manuals/PRRS/PRRS-Version-10.pdf.

Acton, Mark, John Booth, and Daniele Paci. 2025. “2025 Best

Practice Guidelines for the EU Code of Conduct on Data Centre

Energy Efﬁciency.” JRC Publications Repository. https://doi.

org/10.2760/9449356.

Amazon. 2025. “AWS and HUMAIN Announce a More

than $5B Investment to Accelerate AI Adoption

in Saudi Arabia and Globally.” May 13. https://

www.aboutamazon.com/news/company-news/

amazon-aws-humain-ai-investment-in-saudi-arabia.

Aterio. 2025. “Power Capacity Estimation ML Model.” July 2.

https://knowledge.aterio.io/data-products/us-data-centers/

our-modeling-approach/power-capacity-estimation-ml-model.

Bellini, Emiliano. 2021. “Saudi Arabia’s Second PV Tender

Draws World Record Low Bid of $0.0104/kWh.”

Pv Magazine

International

. https://www.pv-magazine.com/2021/04/08/saudi-

arabias-second-pv-tender-draws-world-record-low-bid-of-0104-

kwh/.

Bizo, Daniel. 2023. “Global PUEs – Are They Going Anywhere?”

Uptime Institute Journal.

https://journal.uptimeinstitute.com/

global-pues-are-they-going-anywhere/.

Building and Construction Authority (BCA) and Infocomm

Media Development Authority (IMDA). 2024.

GMDC-2024: BCA-

IMDA Green Mark for Data Centres (Beta Version)

. https://

www1.bca.gov.sg/docs/default-source/docs-corp-buildsg/

sustainability/20241008_gmdc2024_ver1.pdf?sfvrsn=407a219c_0.

Bureau of Experts at the Council of Ministers.

2021. “Personal Data Protection Law.” Laws.Boe.

https://laws.boe.gov.sa/boelaws/laws/lawdetails/

b7cfae89-828e-4994-b167-adaa00e37188/1.

Bureau of Experts at the Council of Ministers. 2022.

“Telecommunications and IT Act.” Laws.Boe. https://

laws.boe.gov.sa/BoeLaws/Laws/LawDetails/

ae610645-e094-48ef-814e-aeb4009d244f/1.

California Department of General Services. 2014. “Requirements

of Data Centers and Server Rooms - 1820.3.” https://www.dgs.

ca.gov/Resources/SAM/TOC/1800/1820-3.

CBRE Group Inc. 2025. 2025 Global Data Center Investor

Intentions Survey. https://mktgdocs.cbre.com/2299/

b889db99-6208-4f7f-97a8-55ad8ac1081f-1999143505/

Global_Data_Center_Investor_In.pdf.

Chinese Ministry of Industry and Information Technology,

Ofﬁce of the Central Cyberspace Affairs Commission, Ministry

of Education, People’s Bank of China, and The State-Owned

Assets Supervision and Administration Commission of the State

Council. 2023.

Action Plan for the High-Quality Development of

Computing Power Infrastructure

. CSET. https://cset.georgetown.

edu/wp-content/uploads/t0573_compute_plan_EN.pdf.

Citi Group. 2024. “Data Center Powerplay: The Chips Have to

Go Somewhere.” May 29. https://www.citigroup.com/global/

insights/data-center-powerplay-the-chips-have-to-go-somewhere.

Clemmons, Elisabeth, and Sean Graham. 2025.

The

Macroeconomic Impacts and Implications of Datacenters: A

Comprehensive Assessment

. IDC. https://my.idc.com/getdoc.

jsp?containerId=US53243025.

Climatiq. 2021. “Emission Factor: Electricity Supplied from Grid

– Saudi Arabia.” https://www.climatiq.io/data/emission-factor/

d642f6c8-32d7-4749-ac48-6463f30e0ae7.

Communications, Space and Technology Commission (CST). 2023.

“Provision of Data Centers Services Regulation.” https://www.cst.

gov.sa/en/regulations-and-licenses/regulations/Document-1546.

Communications, Space and Technology Commission (CST).

2024a.

Annual Report 2024

. https://www.cst.gov.sa/en/

knowledge-center/reports.

Communications, Space and Technology Commission (CST).

2024b. “CST Announces That the ‘Data Center Services

Regulations’ Document Has Entered Into Force.” https://www.

cst.gov.sa/en/media-center/news/CST-Announces-that-the-Data-

Center-Services-Regulations-Document-has-Entered-Into-Force.

Crawford, Kate. 2024. “Generative AI Is Guzzling Water and

Energy.” Nature. https://www.nature.com/articles/d41586-024-

00478-x.pdf.

References

Environment Agency. 2022. “Climate Change Agreements.” GOV.

UK. https://www.gov.uk/guidance/climate-change-agreements--2.

European Commission. 2019.

Commission Regulation (EU)

2019/2020

Ofﬁcial Journal of the European Union

. https://eur-

lex.europa.eu/eli/reg/2019/2020/oj/eng.

European Commission. 2024. “Commission Notice on the

Interpretation and Implementation of Certain Legal Provisions of

the Disclosures Delegated Act.”

Ofﬁcial Journal of the European

Union

. https://eur-lex.europa.eu/eli/C/2024/6691/oj/eng.

European Commission. 2025. “Commission Sets Course for

Europe’s AI Leadership with an Ambitious AI Continent Action

Plan.” https://ec.europa.eu/commission/presscorner/detail/en/

ip_25_1013.

European Development Finance Institutions. 2024.

The EU

Sustainable Finance Rules and Their Implications for Impact

Investors

. https://edﬁ.eu/wp-content/uploads/2024/10/EDFI-

Sustainable-Finance-Mapping-Report-FINAL.pdf.

European Parliament and Council of the European Union. 2022.

Directive (EU) 2022/2464 Of The European Parliament and

of the Council

. European Union. https://eur-lex.europa.eu/eli/

dir/2022/2464/oj/eng.

European Parliament and Council of the European Union. 2023.

Directive (EU) 2023/1791 on Energy Efﬁciency and Amending

Regulation (EU)2023/955 (Recast) (Text with EEA Relevance)

European Union. http://data.europa.eu/eli/dir/2023/1791/oj/eng.

Fioretti, Lapo, Carla La Croce, Andrea Siviero, and Elisabeth

Clemmons. 2024.

The Global Impact of Artiﬁcial Intelligence on

the Economy and Jobs: AI Will Steer 3.5% of GDP in 2030

. IDC.

https://shorturl.at/I2BXc.

Fleck, Anna. 2024. “Which Countries Have the Most Data

Centers?” Statista Daily Data, September 20. https://www.

statista.com/chart/24149/data-centers-per-country.

General Authority for Statistics (GASTAT). 2023.

Electrical

Energy Statistics 2023:

Total Energy Delivered to Grid

Increases by Approximately 5% in 2023.

Riyadh, Saudi

Arabia: General Authority for Statistics. https://stats.gov.sa/

documents/20117/2435281/Electrical+Energy+Statistics+2023+-

+EN.pdf.

Data Center Map. n.d. “Middle East Data Centers – 298 Facilities

from Operators.” https://www.datacentermap.com/middle-east/.

Department for Education and Education and Skills Funding

Agency. 2025. “Streamlined Energy and Carbon Reporting (SECR)

for Academy Trusts.” GOV.UK. https://www.gov.uk/government/

publications/streamlined-energy-and-carbon-reporting-secr-for-

academy-trusts.

Department for Energy Security & Net Zero. 2025.

“Participating in the UK ETS.” GOV.UK. https://www.gov.

uk/government/publications/participating-in-the-uk-ets/

participating-in-the-uk-ets.

Desroches, Clément, Martin Chauvin, Louis Ladan, Caroline

Vateau, Simon Gosset, and Philippe Cordier. 2025. “Exploring

the Sustainable Scaling of AI Dilemma: A Projective Study of

Corporations’ AI Environmental Impacts.” arXiv:2501.14334.

Preprint, arXiv, January 27. https://arxiv.org/abs/2501.14334.

Diamant, Adam. 2022.

24/7 Carbon Free Energy: Matching

Carbon Free Energy Procurement to Hourly Electric Load

Electric Power Research Institute (EPRI). https://restservice.epri.

com/publicdownload/000000003002025290/0/Product.

Donnellan, Douglas. 2023.

Uptime Institute Data Center and

IT Spending Survey 2022

. Uptime Institute. https://datacenter.

uptimeinstitute.com/rs/711-RIA-145/images/DCSpendingSurvey.

Report.01032023.pdf.

Duncan, Glen, Muhd Syaﬁq, Daniel Thorpe, and Kari Beets. 2024.

Data Centers 2024 Global Outlook

. JLL. https://leadsdell.com/

wp-content/uploads/2024/09/jll-data-center-outlook-global-2024.

pdf.

Egbert, Michael. 2025. “Oracle’s Commitment to Saudi Arabia

and President Trump’s Vision for Global Prosperity.”

Oracle

News

. https://www.oracle.com/news/announcement/oracles-

commitment-to-saudi-arabia-and-president-trumps-vision-for-

global-prosperity-2025-05-13/.

Electric Power Research Institute (EPRI). 2024.

Powering

Intelligence: Analyzing Artiﬁcial Intelligence and Data Center

Energy Consumption

. https://www.epri.com/research/

products/3002028905.

England, Andrew, and Ahmed Al Omran. 2025. “Saudi

Arabia Seeks to Use Financial Might to Muscle into Global

AI Industry.”

Financial Times

, May 28. https://www.ft.com/

content/176c7859-fdda-40d2-92a5-15d570f7accf.

References

Groq. 2025. “Saudi Arabia Announces $1.5 Billion Expansion to

Fuel AI-Powered Economy with Groq.” https://groq.com/blog/

saudi-arabia-announces-1-5-billion-expansion-to-fuel-ai-powered-

economy-with-groq.

Hoff, Brandon. 2024. “Semiconductor Spend for Datacenter

Infrastructure and a Breakout for AI Networking.” International

Data Corporation (IDC). https://my.idc.com/getdoc.

jsp?containerId=US52018224.

Houlihan Lokey. 2025.

Real Estate Highlight: Navigating Market

Changes in the Data Center Sector

. https://cdn.hl.com/pdf/2025/

real-estate-highlight-data-centers-march-2025.pdf.

Hui Tian, Yuen. 2024.

Charting Green Growth Pathways at Scale

for Data Centres in Singapore

. Infocomm Media Development

Authority (IMDA). https://isomer-user-content.by.gov.sg/38/

c49f426a-8053-4c7f-bbaa-8437b743a18f/ﬁles-Speeches%202024-

1__Factsheet_IMDA_Green_DC_Roadmap.pdf.

HUMAIN. 2025. “Saudi Arabia’s New AI Enterprise, HUMAIN,

Joins Forces with AMD and Cisco to Launch Groundbreaking AI

Infrastructure Collaboration.” May 13. https://www.humain.ai/en/

news/humain-and-amd-and-cisco/.

IBM. 2025. “What Is an AI Data Center?” https://www.ibm.com/

think/topics/ai-data-center.

Intel. 2025.

Case Study: World’s Largest Green Data Center

https://www.intel.com/content/dam/www/central-libraries/us/

en/documents/2025-04/moro-hub-case-study.pdf.

International Data Corporation. 2024. “IDC’s Worldwide

AI and Generative AI Spending – Industry Outlook

| IDC Blog.” https://blogs.idc.com/2024/08/21/

idcs-worldwide-ai-and-generative-ai-spending-industry-outlook/.

International Energy Agency (IEA). 2024.

Emissions Factors

2024.

https://www.iea.org/data-and-statistics/data-product/

emissions-factors-2024.

International Energy Agency (IEA). 2025a.

Energy and AI

. IEA,

Paris. https://www.iea.org/reports/energy-and-ai.

International Energy Agency (IEA). 2025b.

Building the Future

Transmission Grid

. IEA, Paris. https://www.iea.org/reports/

building-the-future-transmission-grid.

Ghadially, Farokh. 2025. “New Data Centre Rules Could Set a

Global Standard.”

Sustainability Matters Magazine

. https://

www.sustainabilitymatters.net.au/content/sustainability/article/

new-data-centre-rules-could-set-a-global-standard-545515027.

Goldman Sachs. 2024. “AI is Poised to Drive 160%

Increase in Data Center Power Demand.” https://

www.goldmansachs.com/insights/articles/

AI-poised-to-drive-160-increase-in-power-demand.

Google. 2025.

Google Environmental Report 2025

. https://www.

gstatic.com/gumdrop/sustainability/google-2025-environmental-

report.pdf.

Google Cloud. 2025. “Google Cloud and PIF Advance AI Hub

in Saudi Arabia.” Google Cloud Press Corner. https://www.

googlecloudpresscorner.com/2025-05-13-Google-Cloud-and-PIF-

Advance-AI-Hub-in-Saudi-Arabia.

Goverment of Dubai. 2023. “Al Sa’fat – Dubai Green Building

System.” Dubai Municipality. https://www.dm.gov.ae/

municipality-business/al-safat-dubai-green-building-system/.

Grabein, Aaron, and Liz Stine. 2025. “AMD and HUMAIN Form

Strategic, $10B Collaboration to Advance Global AI.” AMD. https://

ir.amd.com/news-events/press-releases/detail/1250/amd-and-

humain-form-strategic-10b-collaboration-to-advance-global-ai.

Graham, Sean, Peter Rutten, and Olga Yashkova. 2024.

Datacenter Capacity, Energy Consumption, and Carbon Emission

Projections

. IDC: The Premier Global Market Intelligence

Company.

Green, Alastair, Humayun Tai, Jesse Noffsinger, Pankaj Sachdeva,

Arjita Bhan, and Raman Sharma. 2024.

How Data Centers and the

Energy Sector Can Sate AI’s Hunger for Power

. McKinsey. https://

www.mckinsey.com/industries/private-capital/our-insights/

how-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-

power.

Greene-Dewasmes, Ginelle, Michael Higgins, and Thapelo

Tladi. 2025.

Artiﬁcial Intelligence’s Energy Paradox: Balancing

Challenges and Opportunities

. World Economic Forum (WEF) in

collaboation with Accenture. https://reports.weforum.org/docs/

WEF_Artiﬁcial_Intelligences_Energy_Paradox_2025.pdf.

References

Istitlaa. 2025. “Global AI Hub Law.” https://istitlaa.ncc.gov.sa/en/

Transportation/citc/globalailaw/Pages/default.aspx.

Jamison, Stephanie, Sanjay Podder, Adam Burden, Bhaskar

Ghosh, Senthil Ramani, Shalabh Kumar Singh, and Matthew

Robinson. 2025.

Powering Sustainable AI

. Accenture. https://

www.accenture.com/content/dam/accenture/ﬁnal/corporate/

corporate-initiatives/sustainability/document/Powering-

Sustainable-AI.pdf.

Kamiya, George, and Vlad C Coroamă. 2025.

Data Centre Energy

Use: Critical Review of Models and Results

. IEA-4E. https://www.

iea-4e.org/wp-content/uploads/2025/05/Data-Centre-Energy-

Use-Critical-Review-of-Models-and-Results.pdf.

King Abdullah Petroleum Studies and Research Center

(KAPSARC). 2023.

Modeling and Projecting Regional Electricity

Demand for Saudi Arabia.

June 5. https://www.kapsarc.org/

our-offerings/publications/modeling-and-projecting-regional-

electricity-demand-for-saudi-arabia/.

King Abdullah Petroleum Studies and Research Center

(KAPSARC). 2025. “Saudi Arabia Renewables Tracker.” https://

apps.kapsarc.org/appboard/renewableprojects.

Kristiansen Nøland, Jonas, Martin Hjelmeland, and Magnus

Korpås. 2024. “Will Energy-Hungry AI Create a Baseload Power

Demand Boom?”

IEEE Access

12: 157824–157836. https://doi.

org/10.1109/ACCESS.2024.3440217.

Levine, Hannah. 2025. “There’s a Labor Shortage at Data Centers.”

JLL. https://www.jll.com/en-us/insights/theres-a-labor-shortage-

at-data-centers.html.

Liou, Joanne. 2023. “What Are Small Modular Reactors (SMRs)?”

IAEA, September 13. https://www.iaea.org/newscenter/news/

what-are-small-modular-reactors-smrs.

Lohn, Andrew, and Micah Musser. 2022.

AI and Compute

. Center

for Security and Emerging Technology (CSET). https://cset.

georgetown.edu/wp-content/uploads/AI-and-Compute-How-

Much-Longer-Can-Computing-Power-Drive-Artiﬁcial-Intelligence-

Progress_v2.pdf.

Malik, Saf. 2025. “What Is Driving Saudi Arabia’s $21bn Data

Centre Investment Surge?” Capacity Media, March 10. https://

www.capacitymedia.com/article/2eir2o1uu11xqfdsdwrgg/news/

what-is-driving-saudi-arabias-21bn-data-centre-investment-surge.

Masanet, Eric, Nuoa Lei, and Jonathan Koomey. 2024. “How

Will the Electricity Use of AI Data Centers Evolve? To Answer

This Question, Energy Analysts Need Better Data.” Preprint,

ResearchGate. http://dx.doi.org/10.13140/RG.2.2.11203.00801.

Mengzhuo, Liu, and Lin Zhewen. 2024. “China’s East Data West

Computing Initiative – Power Infrastructure as the Next Big Thing

in the Global AI Race.” Premia Partners. https://www.premia-

partners.com/sc/insight/china-s-east-data-west-computing-

initiative-power-infrastructure-as-the-next-big-thing-in-the-global-

ai-race.

Meta. 2024.

For a Better Reality: 2024 Sustainability Report

https://sustainability.atmeta.com/wp-content/uploads/2024/08/

Meta-2024-Sustainability-Report.pdf.

Meta. 2025.

Meta’s Odense Data Centre

. Meta. https://

datacenters.atmeta.com/wp-content/uploads/2025/02/Metas-

Odense-Data-Center.pdf.

Ministry of Communications and Information Technology (MCIT).

2020.

KSA Cloud First Policy

. https://www.mcit.gov.sa/sites/

default/ﬁles/cloud_policy_en.pdf.

Ministry of Communications and Information Technology (MCIT).

2023.

Annual Report 2022

. https://www.mcit.gov.sa/sites/

default/ﬁles/2023-07/MCIT_Annual%20Report_2022_En-Web_0.

pdf.

Ministry of Communications and Information Technology (MCIT).

2024.

Annual Report 2024

. https://shorturl.at/73psX.

Mytton, David. 2021. “Data Centre Water Consumption.”

NPJ

Clean Water

4 (1): 11. https://doi.org/10.1038/s41545-021-00101-w.

Narasimhan, Sridhar, Archit Johar, Jay Motani, Santiango

Lavin, Abhilash Jain, and Shiven Mahajan. 2025. “AI

Data Center Location Attractiveness Index.” Kearney.

https://www.kearney.com/industry/technology/article/

ai-data-center-location-attractiveness-index.

National Cybersecurity Authority (NCA). 2018.

National

Cybersecurity Strategy

. https://nca.gov.sa/national_

cybersecurity_strategy-en.pdf.

NEOM. 2025. “DataVolt Signs Agreement with NEOM to Design

and Develop the Region’s First Truly Sustainable, Net-Zero AI

Factory Campus in Oxagon .” https://www.neom.com/en-us/

newsroom/datavolt-signs-agreement-with-neom.

Noffsinger, Jesse, Mark Patel, Pankaj Sachdeva, and Arjita Bhan.

2025.

The Cost of Compute: A $7 Trillion Race to Scale Data

Centers

. McKinsey. https://shorturl.at/H07zW.

NVIDIA Corporation. 2025a. “HUMAIN and NVIDIA Announce

Strategic Partnership to Build AI Factories of the Future in Saudi

Arabia.” NVIDIA Newsroom, May 13. https://nvidianews.nvidia.

com/news/humain-and-nvidia-announce-strategic-partnership-to-

build-ai-factories-of-the-future-in-saudi-arabia.

NVIDIA Corporation. 2025b.

Blackwell Ultra Datasheet

NVIDIA. https://resources.nvidia.com/en-us-gpu-resources/

blackwell-ultra-datasheet.

Ofﬁce of the Federal Chief Sustainability Ofﬁcer. 2025. “Global

Initiatives.” https://www.sustainability.gov/archive/biden46/

federalsustainabilityplan/global-initiatives.html.

Paz, Cassia. 2025. “Texas Senate Bill 6: Data Centers Generator

Fuel Plan Becomes Even More Critical.” Mansﬁeld. https://

mansﬁeld.energy/2025/07/10/texas-senate-bill-6-data-centers-

generator-fuel-plan-becomes-even-more-critical.

Public Investment Fund (PIF). 2024. “PIF and Google Cloud

to Create Advanced AI Hub in Saudi Arabia.” https://www.

pif.gov.sa/en/news-and-insights/press-releases/2024/

pif-and-google-cloud-to-create-advanced-ai-hub-in-saudi-arabia/.

Qualcomm. 2025. “Qualcomm and HUMAIN to Develop State-

of-the-Art AI Data Centers to Deliver Cloud-to-Edge Hybrid

AI Services.” May 13. https://www.qualcomm.com/news/

releases/2025/05/qualcomm-and-humain-to-develop-state-of-

the-art-ai-data-centers-.

Rahman, Robi. 2024. “Leading ML Hardware Becomes 40%

More Energy-Efﬁcient Each Year.” Epoch AI. https://epoch.ai/

data-insights/ml-hardware-energy-efﬁciency.

Saudi & Middle East Green Initiatives. 2025. “SGI Target: Reduce

Carbon Emissions by 278 mtpa by 2030.” https://www.sgi.gov.sa/

about-sgi/sgi-targets/reduce-carbon-emissions.

Saudi Data and AI Authority (SDAIA). 2025a. “National Strategy for

Data & AI.” https://sdaia.gov.sa/en/SDAIA/SdaiaStrategies/Pages/

NationalStrategyForDataAndAI.aspx.

Saudi Data and AI Authority (SDAIA). 2025b. “Saudi Data & AI

Authority and Vision 2030.” https://sdaia.gov.sa/en/SDAIA/

SdaiaStrategies/Pages/sdaiaAnd2030Vision.aspx.

Saudi Electricity Company (SEC). 2025.

Earnings Release

–

2025

. May 14. https://www.se.com.sa/-/media/sec/Investors/

Earning-Reports/earnings-release-q1-2025.ashx.

Saudi Electricity Regulatory Authority (SERA). 2025.

“Consumption Tariff.” https://sera.gov.sa/en/consumer/

electric-tariff/electric-tariff-categories/consumption-tariff.

Saudi Market Research Consulting Firm. 2025. “Saudi AI

Infrastructure Scaling at 30.5% CAGR by 2030.” Saudi

Market Research | Consulting Firm, July 10. https://

saudimarketresearchconsulting.com/insights/articles/

saudi-ai-infrastructure-scaling-at-30-5-cagr-by-2030.

Saudi Press Agency (SPA). 2024. “Minister of Energy Inaugurates

Saudi Smart Grid Conference.” https://www.spa.gov.sa/en/

N2228611.

Saudi Telecom Company (STC). 2024.

Sustainability Report

2024.

Riyadh, Saudi Arabia: stc. https://www.stc.com/content/

dam/groupsites/stc-annual-report-2024/assets/img/pdfs/

sustainability-unsdgs.pdf.

S&P Global Inc. 2025. https://www.marketplace.spglobal.com/en/

datasets.

Sharma, Lakshmee. 2024. “AI Data Centers Threaten Global Water

Security.” Lawfare, December 19. https://www.lawfaremedia.org/

article/ai-data-centers-threaten-global-water-security.

Shehabi, Arman, Dale Sartor, Alex Hubbard, Alex Newkirk, Nuoa

Lei, Md Abu Bakar Siddik, Billie Holecek, Jonathan Koomey, Eric

Masanet, and Dale Sartor. 2024.

2024 United States Data Center

Energy Usage Report

. Lawrence Berkeley National Laboratory.

https://doi.org/10.71468/P1WC7Q.

Shiwani, Amit, Hani Abbasi, and Alistair Levack. 2025. “Unlocking

the Data Centre Opportunity in the Middle East.” PwC. https://

www.pwc.com/m1/en/media-centre/articles/unlocking-the-data-

centre-opportunity-in-the-middle-east.html.

Singla, Alex, Alexander Sukharevsky, Lareina Yee, Michael Chui,

and Bryce Hall. 2025.

The State of AI: Global Survey

. McKinsey

& Company. https://www.mckinsey.com/~/media/mckinsey/

business functions/quantumblack/our insights/the state of

ai/2025/the-state-of-ai-how-organizations-are-rewiring-to-

capture-value_ﬁnal.pdf.

References

Solomon, Steve. 2024. “Sustainable By Design: Next-Generation

Datacenters Consume Zero Water For Cooling.” The Microsoft

Cloud Blog, December 9. Microsoft. https://www.microsoft.com/

en-us/microsoft-cloud/blog/2024/12/09/sustainable-by-design-

next-generation-datacenters-consume-zero-water-for-cooling/.

Spencer, Thomas, and Siddharth Singh. 2024. “What the Data

Centre and AI Boom Could Mean for the Energy Sector.” https://

www.iea.org/commentaries/what-the-data-centre-and-ai-boom-

could-mean-for-the-energy-sector.

Spencer, Thomas, Siddharth Singh, Laura Cozzi, Davide

D’Ambrosio, Hugh Hopewell, Vincent Jacamon, Alex Martinos,

Nicholas Salmon, and Brent Wanner. 2025.

Energy and AI

International Energy Agency (IEA). https://iea.blob.core.

windows.net/assets/601eaec9-ba91-4623-819b-4ded331ec9e8/

EnergyandAI.pdf.

Spindler, Wesley, Luna Atamian Hahn-Petersen, and

Sadaf Hosseini. 2024. “Why Circular Water Solutions

Are Key to Sustainable Data Centres.” World Economic

Forum (WEF). https://www.weforum.org/stories/2024/11/

circular-water-solutions-sustainable-data-centres/.

Springer, Cecilia, and Ali Hasanbeigi. 2025.

Data Centers in the AI

Era: Energy and Emissions Impacts in the U.S. and Key States.

San Francisco, CA: Global Efﬁciency Intelligence, LLC. https://

static1.squarespace.com/static/5877e86f9de4bb8bce72105c/

t/67ab1779a68f9a35968e51f9/1739265952473/

GEI+data+centers+report-+1.15.2025+clean-E.4.pdf.

Srivathsan, Bhargs, Marc Sorel, Pankaj Sachdeva, Arjita Bhan,

Haripreet Batra, Raman Sharma, Rishi Gupta, and Surbhi

Choudhary. 2024.

AI Power: Expanding Data Center Capacity

to Meet Growing Demand

. McKinsey & Company. https://

www.mckinsey.com/industries/technology-media-and-

telecommunications/our-insights/ai-power-expanding-data-

center-capacity-to-meet-growing-demand.

Stansbury, Martin, Kelly Marchese, Kate Hardin, and Carolyn

Amon. 2025. “Can US Infrastructure Keep up with the AI

Economy?” Deloitte. https://www.deloitte.com/us/en/insights/

industry/power-and-utilities/data-center-infrastructure-artiﬁcial-

intelligence.html.

Stokols, Andrew. 2025. “Energy and AI Coordination in the

‘Eastern Data Western Computing’ Plan.” The Jamestown

Foundation, March 8. https://jamestown.substack.com/p/

energy-and-ai-coordination-in-the.

Synergy Research Group. 2025. “The World’s Total Data Center

Capacity Is Shifting Rapidly to HyperscaleOperators.” https://

www.srgresearch.com/articles/the-worlds-total-data-center-

capacity-is-shifting-rapidly-to-hyperscale-operators.

Techusiness. 2025.

AMD Advancing AI 2025 Keynote By CEO

of Humain Tareq Amin

. YouTube. https://www.youtube.com/

watch?v=8UG92qu0Krg.

Tohme, Hani, Adel Belcaid, Lukas de Sonnaville, Arianna Molino,

Rita Carvalho, and Hale De Vera. 2025. “A More Regenerative

Digital Age: The Middle East Data Center Opportunity.” Kearney.

https://www.kearney.com/service/sustainability/article/a-

greener-digital-age-the-middle-east-data-center-opportunity.

Tonomus. 2025.

Compute.

https://tonomus.neom.com/en-us/

what-we-do/compute.html.

U.S. Department of Commerce. 2025. “UAE and US Presidents

Attend the Unveiling of Phase 1 of New 5GW AI Campus

in Abu Dhabi.” https://www.commerce.gov/news/press-

releases/2025/05/uae-and-us-presidents-attend-unveiling-phase-

1-new-5gw-ai-campus-abu.

U.S. Department of Energy, Ofﬁce of Energy Efﬁciency and

Renewable Energy (EERE). 2014.

Early Markets: Fuel Cells for

Backup Power

. https://www.energy.gov/eere/fuelcells/articles/

early-markets-fuel-cells-backup-power.

U.S. Environmental Protection Agency (EPA). 2024.

Inventory of U.S. Greenhouse Gas Emissions and Sinks:

1990–2022

. EPA 430-R-24-004. U.S. Environmental

Protection Agency. https://www.epa.gov/ghgemissions/

inventory-us-greenhouse-gas-emissions-and-sinks-1990-2022.

Van Zandt, Brooke. 2023. “A Decade of Greener Computing

Blooms Inside NREL’s Data Center.” National Renewable Energy

Laboratory (NREL) News & Feature Stories, August 2. https://

www.nrel.gov/news/detail/program/2023/a-decade-of-greener-

computing-blooms-inside-nrels-data-center.

Wang, James. 2024. “Cerebras CS-3 vs. Nvidia B200: 2024 AI

Accelerators Compared.” Cerebras, April 12. https://www.

cerebras.ai/blog/cerebras-cs-3-vs-nvidia-b200-2024-

ai-accelerators-compared.

The White House. 2025.

Americas AI Action Plan

. https://www.

whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-

Action-Plan.pdf.

References

Xinhua. 2024. “China Sets Green Targets for Data Centers.”

The State Council of the People’s Republic of China.

https://english.www.gov.cn/news/202407/24/content_

WS66a0b167c6d0868f4e8e96ba.html.

Xu, Jing, Liping Li, Bin Zhang, Mingsong Hao, and Fengqiao Mei.

2023. “‘Dual Carbon Goals’ Enhances Policy Integration: Analysing

Recent Changes in China’s National Climate Policy.”

Journal of

Asian Public Policy

18(3): 1–18. https://doi.org/10.1080/17516234.2

024.2305320.

Yeo, Amelia. 2022. “Singapore New Data Centers.” The

International Trade Administration, August 11. https://www.trade.

gov/market-intelligence/singapore-new-data-centers.

Zhang, Ning, Huabo Duan, Yuru Guan, Ruichang Mao, Guanghan

Song, Jiakuan Yang, and Yuli Shan. 2024. “The ‘Eastern Data and

Western Computing’ Initiative in China Contributes to Its Net-

Zero Target.”

Engineering

52: 256–261. https://doi.org/10.1016/j.

eng.2024.08.010.

References

Acknowledgments

The authors would like to thank Reema Alnuwisr, whose contributions to the cost

analysis, coding, and data analytics were helpful to this study. The authors also wish to

express their appreciation to the copyediting and design team at KAPSARC, especially

Bree De Roche, Syed Yunus, and Christopher Bartle, for their editorial support and

assistance in preparing the ﬁnal publication.

Appendices

Appendix A. Scenario-Based

Framework for Estimating

Data Center Capacity

(2025–2030)

We developed three growth scenarios to account for uncertainty

in the trajectory of Saudi Arabia’s data center expansion. These

estimates are mainly derived from the S&P Global datasets and

supplemented by information from announced projects.

Growth Scenarios

• Baseline (non-AI growth): Based on planned non-AI

facilities, capacity expands from 290.5 MW in 2024 to about

1,050 MW by 2030.

• Moderate AI growth: Includes confirmed AI projects such

as HUMAIN-AI Factory 1, Dammam (250 MW), HUMAIN-

AI Factory 2 (250 MW), Datavolt-Riyadh (150 MW), and

Datavolt-NEOM (300 MW). Adding these to the baseline

scenario yields around 2,000 MW by 2030.

• High AI growth: Assumes that all announced mega-projects

proceed, including HUMAIN’s 1,900 MW target by 2030 and

NEOM’s AI campus (1,150 MW by 2030 of its larger 1,500

MW). With the baseline of 1,050 MW included, total capacity

reaches about 4,100 MW by 2030.

The operational conditions were based on the study “Data

Centers in the AI Era,”13 which has two operational conditions:

a “business-as-usual” scenario where data centers operate with

conventional efﬁciency, and an “advanced efﬁciency” scenario

where new technologies and best practices are widely adopted.

The interaction of the capacity growth scenarios and the

operational conditions gives us the following intersection:

• High growth, conventional: Rapid expansion to 4,100

MW by 2030, prioritizing speed over sustainability. Limited

adoption of efficiency measures. This scenario produces the

highest energy use and CO2 emissions, serving as a stress

test for power and resource systems.

• High growth, sustainable: Also achieves 4,100 MW by 2030

but integrates energy efficiency with expansion and scales

up renewable energy, reducing electricity consumption

and CO2 emissions significantly due to higher operational

efficiency.

• Moderate growth, conventional: Capacity increases

steadily to around 2,000 MW with a conventional operating

model. Efficiency is limited, leading to some increases

in energy use and emissions, and does not align with

sustainability goals.

• Moderate growth, sustainable: Capacity increases to 2,000

MW but with strong efficiency gains and alignment with

national green targets. PUE improves through adopting best

practices, and renewable integration increases, lowering

both electricity consumption and CO2 emissions compared

to the conventional scenarios.

13 Springer and Hasanbeigi (2025).

total

= [(

) + (

non-AI

)]

/1000

(2)

Where

•

non-AI

are the installed IT capacity in MW for AI and non-

AI facilities, respectively.

•

, u

non-AI

are the utilization rates for AI and non-AI facilities,

respectively.

•

non-AI

are the assumed PUE values for AI and non-AI

facilities, respectively.

•

is the annual operating hours (8,760).

Estimating the Total Electricity Demand

by 2030

To estimate the share of total electricity demand by data centers

in Saudi Arabia, we used existing projections and reported

statistics of national electricity consumption. Projections from

the King Abdullah Petroleum Studies and Research Center

(KAPSARC) indicate that demand could reach 365,400 GWh

by 2030.15 under a scenario of moderate economic growth and

stable prices. For the years 2022 and 2023, we used actual

electricity consumption ﬁgures published by the General

Authority for Statistics (GASTAT), which were 309,524 GWh and

327,001 GWh, respectively.16 We derived values for the years

2024 to 2029 by a linear interpolation between the ofﬁcial

statistics and the KAPSARC projection, giving us a consistent

trajectory toward the 2030 estimate.

Appendix B. Methodology

for Estimating Data Center

Energy Demand in Saudi

Arabia (2025-2030)

Estimating Data Centers’ Energy Demand

This study estimates the electricity demand of data centers in

Saudi Arabia for 2025-2030 using the framework proposed by

Masanet et al. (2024)14 rather than detailed bottom-up server-

or rack-level measurements. Our method estimates total

energy use based on installed IT capacity, adjusted by usage

rate, PUE, and annual operating hours. The general formula

of the estimated data center electricity use in

total

GWh is

expressed as

total

1000 (1)

where

is the critical IT capacity in MW,

is the average of

critical IT capacity used,

is the power usage effectiveness, and

is the annual operating hours (8,760).

Because operational patterns differ between AI and general-

purpose facilities, we adopted a differentiated formulation of

the electricity consumption equation, in which the PUE and

usage rates are assigned according to workload type. The

extended equation is expressed as:

Table B1. Projected growth in data center electricity demand and its share of total national electricity consumption.

Capacity (MW) Energy consumption

(TWh/year)

Share of national

electricity demand

Current (2024) 290.5 2.80 0.85%

Scenario 0: Non-AI growth 1,050 10.16 2.79%

Scenario 1: Moderate growth, conventional 2,000 20.15 5.52%

Scenario 2: Moderate growth, sustainable 2,000 17.62 4.8%

Scenario 3: High growth, conventional 4,100 42.23 11.55%

Scenario 4: High growth, sustainable 4,100 36.76 10.1%

Appendices

14 Masanet, Lei, and Koomey (2024).

15 KAPSARC (2023).

16 GASTAT (2023).

Appendices

Key Operational Assumptions: PUE and

Utilization Rate

PUE and utilization rate are two critical variables in data center

energy modeling. In Saudi Arabia, data centers record PUE values

between 1.80 and 2.1017 due to the hot climate, though recent

data reports a national average closer to 1.53.18 New centers,

such as NEOM’s Oxagon campus, aim for values below 1.3 in

hyperscale facilities,19 actively pursuing efﬁciency improvements.

To investigate the role of efﬁciency on electricity demand, we

looked at two distinct operational conditions: conventional

and sustainable. Under the conventional conditions in our

study, current general-purpose data centers using air-cooled

designs are the least energy efﬁcient, with a PUE of around

1.70, consistent with global benchmarks for infrastructure in hot

climates.20 AI-oriented centers were given a moderately lower

PUE of 1.50, reﬂecting the adoption of liquid cooling but with

limited system-wide optimization.

In the sustainable condition scenario, we assumed that

improvements in efﬁciency will increase over time,

Table B2. Assumed PUE and utilization rate values for different operational conditions and

data center types in Saudi Arabia by 2030.

Scenario Data center type PUE (2030) Utilization rate (2030)

Conventional General purpose 1.70 0.65

AI 1.50 0.80

Sustainable General purpose 1.50 0.65

AI 1.30 0.80

consistent with global trends in advanced cooling and energy

management. Modern general-purpose centers are expected

to achieve a PUE of around 1.5 while AI-optimized centers with

the latest in cooling technology could reach values near 1.30. As

shown in Table B2, these assumptions recognize that efﬁciency

proﬁles differ between AI-intensive and general-purpose data

centers, due to variations in their workload density and cooling

requirements.

The utilization rates are another key factor in modeling

electricity demand. AI-focused data centers, particularly

those supporting large-scale model training and continuous

inference, have a higher utilization rate due to long-duration,

compute-intensive workloads with minimal idle periods.

Recent studies report utilization rates of around 75%-85%

for AI training workloads, with lower values for inference.21

Enterprise and colocation facilities typically have more

variable workloads, resulting in a lower average utilization,

commonly in the range of 50%-70%.22 For this study, AI

facilities were given a utilization rate of 0.80, and non-AI

facilities a rate of 0.65.

17 STC (2024).

18 S&P Global dataset.

19 Tonomus (2025).

20 Bizo (2023).

21 Shehabi et al. (2024).

22 Citi Group (2024).

Appendix C. Methodology

for Estimating CO2 Emissions

from Data Centers’ Electricity

Consumption

Estimating CO2 Emissions

To quantify the carbon footprint of Saudi Arabia’s data centers,

we linked projected electricity consumption to the national

energy mix using standardized emissions factor methods from

the IEA CO2 Emissions Factors 2024 dataset.23 The estimation

follows the fundamental equation

Emissions

(

) =

total

× 10-3 (3)

Where

•

total

is the total annual energy demand of data centers

(

GWh

•

is the grid emission factor, which represents the average

CO2 emitted per MWh of electricity generated, reflecting the

national energy mix (

tCO

MWh

The grid emission factor for Saudi Arabia varies based on the

energy mix:

• Fossil fuel mix: (41.2% oil, 58.2% natural gas,

<1% renewables) results in a grid emission factor24 of

0.568 tCO2/MWh25:

= (0.412 × 0.783

tCO

MWh

) + (0.582 × 0.423

tCO

MWh

)

= 0.568 tCO2/MWh

• Vision 2030 target mix: (50% natural gas, 50% renewables)

reduces the grid emission factor to 0.21 tCO2/MWh:

= (0.5 × 0.42

tCO

MWh

) + (0.5 × 0

O2/

MWh

)

= 0.21 tCO2/MWh

Estimating CUE

The carbon usage effectiveness (CUE) in tCO2/MWh-IT is

calculated as the product of PUE (

) and the grid emission

factor (

CUE

= α ×

(4)

Using the following grid factors and PUE ranges:

• Conventional (fossil fuel mix 0.568 tCO2/MWh; PUE 1.5-1.7):

CUE = 0.852-0.966 tCO2/MWh-IT.

• Sustainable (Vision 2030 mix 0.21 tCO2/MWh; PUE 1.3-1.5):

CUE = 0.273-0.315 tCO2/MWh-IT.

23 IEA (2024).

24 Greenhouse Gas Emissions Inventory (EPA 2024).

25 Climatiq (2021).

Appendices

Table C1. Summary of data center growth scenarios in Saudi Arabia from present to 2030.

Scenario

Variable Current

(2024)

Scenario 0

Non-AI growth

Scenario 1

High growth,

conventional

Scenario 2

High growth,

sustainable

Scenario 3

Moderate growth,

conventional

Scenario 4

Moderate growth,

sustainable

IT load capacity (MW) 290.5 MW 1,050 MW Non-AI: 1,050 MW

AI: 3,050 MW = 4,100 MW

Non-AI: 1,050 MW

AI: 950 MW = 2,000 MW

Workload type General purpose and

enterprise

Predominantly AI model training

and inference

Mixed: general purpose and AI

PUE 1.7 1.7 AI: 1.5

Non-AI: 1.7

AI: 1.3

Non-AI: 1.5

AI: 1.5

Non-AI: 1.7

AI: 1.3

Non-AI: 1.5

Renewable share (%) 1% 1% 1% 50% 1% 50%

Usage rate 0.65 0.65 AI: 0.80

Non-AI: 0.65

AI: 0.80

Non-AI: 0.65

AI: 0.80

Non-AI: 0.65

AI: 0.80

Non-AI: 0.65

Derived

metrics

Total electricity demand (TWh) 2.80 10.16 42.23 36.76 20.15 17.62

Share of national electricity

demand (%)

0.85% 2.79% 11.55% 10.10% 5.52% 4.80%

CO2 emissions (MtCO2/year) 1.60 5.81 24.02 7.7 11.48 3.7

CUE (tCO2/MWh-IT) 0.966 0.966 0.852-0.966 0.273-0.315 0.852-0.966 0.273-0.315

Note: Scenarios vary by capacity expansion and operational conditions. The derived metrics are the total electricity demand, share of national demand, and CO

emissions.

Appendices

• Appendix D. Methodology

for Estimating Lifetime Data

Center Project Costs

The cost analysis needs to consider the PUE, load factor,

computing efﬁciency, electricity price, CAPEX, and WACC.

Hence, its formula needs to be a consolidated, time-discounted

view of the total costs incurred per unit of computing output

over the operational lifetime of a data center. The unit is

($/PFLOP) and its formula is



αpkt

: the energy costs, where p is the electricity

price in ($/MWh), scaled by the load factor k and

the power use effectiveness α, and multiplied by

the average time per year t



: Non-energy OPEX. Here, d is a percentage of the

CAPEX c.



∑

-1

(

)

: This is to discount each year for the

time horizon of interest (project lifetime). The

discount factor r is the weighted average cost of

capital (WACC).

• Denominator (PFLOPS/kW):

◦γ: The computing efficiency of the AI hardware in

(PFLOPS/kW)

◦

∑

-1

(

)

: The effective compute output each year,

discounted over time

Formula (5) is quite intuitive. The more efﬁcient the AI hardware

is, the lower the cost, reﬂecting incentives in using energy-

efﬁcient hardware. This is also consistent with the trend seen

in Table 2, which shows signiﬁcant improvements over the past

few years. The electricity price and the PUE directly affect costs

via OPEX as major contributors to the energy-related costs. One

can easily see that the cost is linear in the price

and the PUE

α, but nonlinear in the computing efﬁciency γ as it is scaled

by (1/γ). It is also nonlinear in the load factor

, which directly

affects OPEX and compute output.

(5)

∑

-1

αpkt

(

)

-1

(

)

CAPEX

OPEX

Present Value of

Total Output over

Project Lifetime

DC Project Cost

= =

At a high level, the cost structure mirrors the levelized cost of

energy, with the output changed from electricity generation

to computational throughput. The numerator represents the

total discounted costs of the data center, and the denominator

represents the total discounted compute output. We discuss

each term individually below:

• Numerator ($/kW):

◦c: initial overnight cost of construction (CAPEX) per kW

of IT load

◦

∑

-1

αpkt

(

)

: discounted operating expenses (OPEX)

for the lifetime of the data center

Appendices

Appendix E. Glossary of Definitions

Term Deﬁnition

Artiﬁcial intelligence (AI) A computer science ﬁeld that focuses on building systems capable of performing tasks that usually require human

intelligence, such as learning, reasoning, and self-development.

AI-specialized data center A facility designed speciﬁcally for AI workloads, typically featuring high-performance GPUs, advanced CPUs, and

specialized infrastructure to support dense, compute-heavy operations.

AI workload A category of computational tasks that includes machine learning, deep learning, generative AI, and other

resource-intensive AI applications.

Carbon usage effectiveness (CUE) A standard metric measuring the carbon emissions intensity per MWh delivered to IT equipment (tCO2/MWh-IT).

Central processing unit (CPU) The main processing chip in a computer that executes general-purpose tasks and controls other components.

Compound annual growth rate

(CAGR)

The average annual growth rate of a value over a speciﬁed period, assuming it grows at a steady rate each year.

Compute (in AI context) Refers to the computational power, typically measured in FLOPS, required to train or run AI models. In the context

of data centers, compute includes the combined processing capacity of specialized hardware such as GPUs, TPUs,

and AI accelerators, which are essential for executing complex machine learning tasks at scale.

Data center A physical facility that houses many servers and data storage devices with high-speed connectivity to manage an

organization’s applications and data.

Data center infrastructure efﬁciency

(DCIE)

A measure of a data center’s energy efﬁciency calculated as the percentage of energy used directly by IT

equipment out of the total energy consumption. Higher DCIE values signify greater efﬁciency in non-computational

functions.

FLOPs (ﬂoating point operations) The total count of ﬂoating-point operations performed to complete a speciﬁc task or program.

GPU (graphics processing unit) A specialized processor designed for parallel operations, originally for rendering graphics but now widely used in

AI and high-performance computing due to its ability to handle complex, large-scale computations efﬁciently.

Inference A process for making predictions by applying a trained model to unlabeled examples.

IT load capacity The maximum power demand (measured in megawatts) of servers and network equipment installed in a data

center, excluding cooling and ancillary systems.

The Jevons Paradox An economic principle stating that improvements in resource efﬁciency can paradoxically lead to increased overall

consumption of that resource, due to reduced costs and rising demand.

Language processing unit (LPU) A specialized chip designed to optimize natural language processing tasks such as real-time translations, speech

recognition, and text analysis. LPUs are more energy-efﬁcient for speciﬁc AI applications compared to general-

purpose processors.

Model training The process of determining or optimizing the parameters of a model based on a machine learning algorithm using

training data.

Power usage effectiveness (PUE) A metric for data center energy efﬁciency, calculated as the ratio of total facility power to computing equipment

power. A PUE closer to 1.0 indicates higher efﬁciency.

Rack power density The amount of power consumed per server rack, typically measured in kilowatts. Higher densities are common in

AI and high-performance computing environments.

Utilization rate The proportion of a data center’s installed IT capacity that is actively used. Higher utilization rates indicate more

efﬁcient use of infrastructure.

Workload A combination of tasks that run on a given computer system.

Appendices

Khaled Alshehri

Khaled Alshehri is a Research Fellow in the Utilities and Renewables Program at the

King Abdullah Petroleum Studies and Research Center (KAPSARC). His research

focuses on integrating digital technologies into the power sector to optimize

grid operations, enhance economic efﬁciency, and support the energy transition.

Drawing on extensive experience across government, academia, and industry,

Dr. Alshehri leverages advanced tools from game theory, control, and optimization

to address complex challenges in energy systems. In 2024, Dr. Alshehri was honored

with the Questrom-CEMA Best Paper Award for his co-authored work on the

efﬁcient aggregation of distributed energy resources – a recognition of signiﬁcant

contributions to the ﬁeld. He earned his B.S. in Control and Instrumentation

Engineering from King Fahd University of Petroleum and Minerals (KFUPM) and his

M.S. and Ph.D. in Electrical and Computer Engineering from the University of Illinois at

Urbana-Champaign (UIUC).

Marwa Khan AlFattani

Marwa Khan AlFattani is a Researcher and Technology Strategist with more than 15

years of experience specializing in artiﬁcial intelligence, digital transformation, and

national technology studies. In her Strategic leadership roles, she has led research

initiatives that bridge AI research, policy development, and strategic implementation,

advancing the Kingdom’s digital transformation agenda. Previously, AlFattani

spent nearly a decade at King Abdulaziz City for Science and Technology (KACST),

contributing to Saudi Arabia’s ﬁrst AI governance report and conducting applied

research across machine learning, natural language processing, simulation systems,

and smart home technologies. She is an inventor with a registered patent and has

authored peer-reviewed publications in the ﬁelds of AI, digital identity, and Arabic

language technologies. AlFattani holds a Master of Science in Information Technology

from King Saud University and continues to focus on responsible AI and public-sector

innovation.

About the Authors

AI and Energy: The Future of Data Centers in Saudi Arabia

Laila Bashmal

Laila Bashmal is an Artiﬁcial Intelligence Researcher with a Ph.D. in Computer

Engineering from King Saud University. Her work focuses on leveraging uniﬁed AI

systems for high-impact applications, particularly in remote sensing analysis and

medical image processing. Her broader research interests include large language

models, multimodal learning, and the integration of language and vision to build more

general and robust AI systems, as well as examining how these capabilities reshape

computational infrastructure and real-world applications.

Ghaliah Alshammari

Ghaliah Alshammari is a Senior Researcher, specializing in data science, artiﬁcial

intelligence, and analytical innovation. Her work focuses on developing data-driven

insights that support national digital transformation. She previously served as an

Information Technology Analyst with the G20 Saudi Secretariat. Ghaliah holds an

M.S. in Computer Science from The City College of New York and a B.S. in Computer

Science from Princess Nourah Bint Abdulrahman University.

About the Project

This discussion paper is part of the “Quantifying the Value of Generative AI (GenAI) for

the Saudi Power Sector” project. Recognizing that digital infrastructure is a prerequisite

for deploying GenAI at scale, the paper provides a ﬁrst comprehensive assessment of

the landscape, energy demand, emissions proﬁle, and cost dynamics of AI-oriented data

centers in Saudi Arabia. It outlines key scenarios, operational considerations, and strategic

enablers that will shape the Kingdom’s capacity to support emerging GenAI applications.

70AI and Energy: The Future of Data Centers in Saudi Arabia

kapsarc.org /kapsarcinfo@kapsarc.org