UAE AI Data Center Timeline 2017–2026: Inside the 5-Gigawatt AI Infrastructure Plan
How the UAE went from a 2017 AI strategy to a 5-gigawatt AI campus with G42, OpenAI and Microsoft — and why 2026’s regional attacks are reshaping where
Latest Story
Nine years ago, the UAE decided artificial intelligence would be a national project, not just a technology purchase. That bet is now a physical one: alongside G42 and a group of American partners — OpenAI, Oracle, Nvidia, Cisco and SoftBank — the UAE has announced a 5-gigawatt AI campus in Abu Dhabi, one of the largest AI infrastructure projects planned anywhere outside the United States. Five gigawatts is a measure of electrical power, not intelligence — roughly the combined output of several large power stations, dedicated to running servers, chips and cooling systems. Building it means the UAE, Microsoft, OpenAI and Nvidia are now as involved in power plants, water-cooling and fiber routes as in AI models. And in 2026, that physical reality collided with regional conflict: Iranian strikes reportedly knocked Amazon data centers offline in the UAE and Bahrain, and by September, Reuters reported that UAE planners were rethinking whether to concentrate so much AI infrastructure in one place at all.
🧠 What Is the UAE’s 5-GW AI Data-Center Project?
The UAE, G42 and U.S. partners including OpenAI, Oracle, Nvidia, Cisco and SoftBank have announced a 5-gigawatt UAE-U.S. AI Campus near Abu Dhabi — a planned site of roughly 10 square miles meant to eventually host 5 GW of power capacity for AI computing. 5 GW of planned infrastructure capacity is not the same as 5 GW of computing performance — it describes how much electricity the site is designed to draw, not how “smart” its AI is. Inside that campus, a first cluster called Stargate UAE (1 GW, built by G42, run by OpenAI and Oracle) is under construction; only an initial 200-megawatt phase was targeted for completion by Q3 2026. The rest of the 5 GW remains a plan, not a built facility.
Fast Answers Before You Read On
What to Remember
- 5 GW measures power, not intelligence. It describes electrical capacity the campus is designed to eventually draw — not GPU count, model quality or “how smart” the AI is.
- Stargate UAE is a piece of the 5-GW campus, not the whole thing. It is a 1-GW compute cluster within the larger site — and only a 200-MW first phase was targeted for 2026.
- Most of the 5 GW does not exist yet. The campus is real as an announced, partly-under-construction project — not as finished, operating infrastructure.
- The UAE’s AI push has an 8-year paper trail: a 2017 national strategy, a 2019 graduate university, a homegrown AI champion (G42) since 2018, and a fast-deepening U.S. technology partnership since 2024.
- Chips became diplomacy. Access to Nvidia’s most advanced accelerators required a U.S. government security framework, not just a purchase order — UAE-approved entities only got a license-free path in July 2026.
- 2026 added a hard lesson in physical risk. Data centers, described casually as “the cloud,” turned out to be physical buildings that Iranian strikes reportedly could and did knock offline elsewhere in the Gulf.
- Reported plan changes are not yet decisions. As of this update, distributing sites, hardening or building underground are options reportedly under consideration, per Reuters sourcing — not an announced final design.
- Hydrocarbons are not being replaced. The honest framing is economic diversification: AI and data-center infrastructure are becoming an additional strategic asset alongside oil and gas, not a substitute for them.
Interactive: Can You Protect a 5-GW AI City?
Tap each system a real AI campus depends on — then see what a regional disruption exposes
🏜️ Abu Dhabi, pre-construction. Building a 5-GW AI campus means solving four problems at once — power, cooling, compute and connectivity — before a single model ever runs.
⚡ Power — How Much, From Where?
The campus is designed to scale in increments toward 5 GW total. G42’s own announcement says the site will eventually draw on nuclear, solar and natural gas power to limit emissions — a stated goal for the finished campus, not a confirmed fuel mix for every megawatt already built.
❄️ Cooling — Air, Water or Hybrid?
No cooling method wins universally in a desert climate. Air cooling is simplest but least efficient at scale; chilled-water and hybrid systems handle dense AI racks better but need a reliable water source; direct-to-chip liquid cooling is increasingly used for the newest accelerators. The right choice depends on rack density, local water access and cost — not a single “best” answer.
🧠 Compute — Chips, Servers, Storage
This is the layer people picture when they hear “AI data center”: AI accelerators (like Nvidia’s Blackwell-class GPUs), the servers that hold them, and storage for training data and model weights. It is also the layer most exposed to export-control policy — see the chips section below.
🌐 Connectivity — Fiber and Redundant Routes
None of this compute is useful in isolation. It needs high-speed fiber to reach users, cloud regions and other data centers, plus more than one physical route in and out — so a single cut cable or damaged junction doesn’t isolate the whole site.
🔒 Resilience — Backup and Redundancy
Backup generators, battery buffers, redundant network paths and (increasingly, per 2026 reporting) distributed capacity across more than one site all reduce how much a single failure can take down.
🚨 Power Grid
A single-site campus depends on the local grid and its own on-site generation holding up together — a regional grid disruption threatens the whole cluster at once.
🚨 Fiber
Fiber cuts or damaged network junctions near one physical site can isolate an entire campus’s compute from the internet, even if the servers themselves are untouched.
🚨 Cooling
Chillers, cooling towers and water intakes are typically less hardened than the server halls they serve — disabling cooling alone can force a shutdown without a direct hit on any computer.
🚨 Chip Supply
Advanced accelerators depend on a small number of manufacturers and an export-approval chain; a policy change or supply disruption can stall expansion even if the building is intact.
🚨 Physical Location
Concentrating gigawatts of compute in one place concentrates risk in one place — the logic behind reported UAE interest in distributing sites.
🚨 Water & Environment
Water-cooled systems need a reliable, sustainable water supply in a desert environment — a genuine engineering constraint, not an afterthought.
That is the real lesson of 2026: would you still put everything in one place?
The Number: What Does 5 GW Actually Mean?
A gigawatt measures power — not AI intelligence, GPU count or computing performance
A gigawatt is one billion watts — a unit of electrical power, the same unit used for power plants and national grids. One gigawatt is roughly the continuous output of a large power-station unit. Five gigawatts, if ever fully built and running, would be a serious claim on a national grid — but it says nothing on its own about how capable the AI running on it is, how many chips sit inside it, or how “intelligent” any model trained there might be. This article does not convert 5 GW into a GPU count: that number depends on chip generation, power-per-rack design and efficiency assumptions that have not been officially disclosed for this specific campus.
Announced / Planned Capacity
The campus’s stated end-state ambition, announced May 2025. This is a target, confirmed as an announcement — not as completed construction.
Reported Under-Construction Capacity
The first phase of Stargate UAE’s 1-GW cluster, targeted for completion by Q3 2026 — about 4% of the full 5-GW plan, and the only slice with a public construction timeline.
UAE AI Timeline: 2017–2026
Newest first — every date checked against primary announcements or Reuters/AP-level reporting
Reuters: UAE Reconsiders Concentrating AI Infrastructure in One Place
What happened: On 11 September 2026, Reuters reported, citing people familiar with the discussions, that the UAE is revising its 5-GW plan after the Iran war. The originally envisaged single 10-square-mile Abu Dhabi campus would likely become a network of sites spread across the UAE, with underground construction, blast-resistant structures and stronger drone/missile defenses among the options under review.
UAE Pledges €40 Billion to Germany, Including AI and Energy
What happened: During President Sheikh Mohamed bin Zayed’s 9–11 September state visit to Germany, the UAE pledged €40 billion across industry, AI, digital infrastructure and energy. ADNOC, XRG and Masdar separately signed deals with German energy firms covering LNG, renewables and hydrogen. This is a distinct diversification and diplomacy story — not part of the 5-GW Abu Dhabi campus.
UAE Upgraded to Top U.S. Chip-Export Tier
What happened: On 10 July 2026, the U.S. Commerce Department’s Bureau of Industry and Security moved the UAE into Country Group A:5, its highest export-control tier, removing it from the more restrictive D:3/D:4 groups. This allows the UAE government and approved entities such as G42 — plus U.S. firms including Amazon, Microsoft, Google, Meta, OpenAI, Oracle and xAI operating in the UAE — to import advanced AI chips and servers without individual export licenses.
Iranian Strikes Reportedly Knock Three AWS Data Centers Offline
What happened: Around 1 March 2026, drones or missiles reportedly struck three Amazon Web Services data centers — two in the UAE, one in Bahrain — forcing them offline and disrupting banking, payments, delivery apps and enterprise software across the region. AWS declined to comment; Iran’s Fars News Agency claimed the sites supported military and intelligence activity. Neither UAE nor Bahraini authorities are on record confirming the details.
G42 Targets Q3 2026 for Stargate UAE’s First 200 MW
What happened: G42 subsidiary Khazna Data Centers provided a construction update on 5 December 2025: the first 200 megawatts of Stargate UAE’s planned 1-gigawatt build-out were targeted for completion in the third quarter of 2026, on what the company called an accelerated timeline, with full 1-GW build-out targeted within roughly three years.
U.S. Approves Advanced Nvidia Chip Exports to G42
What happened: On 19–20 November 2025, the U.S. government approved exports of Nvidia’s advanced Blackwell-class (GB300) AI chips to G42 in the UAE and Humain in Saudi Arabia — up to the equivalent of 35,000 chips each — under security and reporting requirements tied to the AI Acceleration Partnership framework.
Stargate UAE Launches Inside the New 5-GW Campus
What happened: On 22 May 2025, G42 launched Stargate UAE with OpenAI, Oracle, Nvidia, Cisco and SoftBank — a 1-gigawatt AI compute cluster to be built by G42 (via Khazna) and operated by OpenAI and Oracle, positioned as the largest AI campus project outside the United States. This followed the 5-GW UAE-U.S. AI Campus announcement days earlier, at Qasr Al Watan, in the presence of Sheikh Mohamed bin Zayed and President Trump, during Trump’s UAE state visit.
MGX Launches and UAE Pledges $1.4 Trillion U.S. Investment
What happened: MGX Fund Management, an Abu Dhabi state-backed AI investment vehicle founded by Mubadala and G42 with a stated ambition to manage $100 billion in assets, launched in March 2025. That same month, following a meeting between President Trump and UAE national security adviser Sheikh Tahnoon bin Zayed, the UAE pledged a 10-year, $1.4 trillion investment framework in the U.S. spanning AI infrastructure, semiconductors, energy and manufacturing.
Microsoft Invests $1.5 Billion in G42
What happened: On 16 April 2024, Microsoft announced a $1.5 billion minority investment in G42, with Microsoft Vice Chair and President Brad Smith joining G42’s board. The deal included a first-of-its-kind Intergovernmental Assurance Agreement between the U.S. and UAE governments on AI safety and security standards, and had G42 run its AI applications on Microsoft Azure.
Generative AI Boom Makes Compute a Strategic Resource
What happened: As ChatGPT-driven demand reshaped global AI investment, G42’s Inception unit released Jais on 30 August 2023 — described as the world’s most advanced open-source Arabic large language model, trained on the Condor Galaxy 1 supercomputer built with Cerebras, in collaboration with MBZUAI. Globally, GPU access, cloud capacity and electricity supply all became strategically important almost overnight.
Cloud and AI Infrastructure Expand: MBZUAI, Artemis, Khazna
What happened: MBZUAI — the world’s first graduate-level, research-focused AI university — was initiated in October 2019, granted a university license in April 2020, and began teaching graduate students in January 2021. In parallel, G42’s Artemis supercomputer (built with Nvidia) ranked #26 globally on the TOP500 list in November 2019, the first time a UAE-based organization cracked the world’s top 100. Data-center capacity, led by G42 subsidiary Khazna, expanded through this period to support the growing ecosystem.
G42 Is Founded in Abu Dhabi
What happened: Group 42 (G42) was founded in 2018, chaired by Sheikh Tahnoon bin Zayed Al Nahyan — the UAE’s national security adviser — and led by CEO Peng Xiao, with a stated mission to make AI “a national utility, much like electricity or water.” It became a unicorn by 2021 and grew into the UAE’s central AI technology group.
UAE Launches Its National AI Strategy
What happened: In October 2017, the UAE government launched the UAE Strategy for Artificial Intelligence, part of the wider UAE Centennial 2071 vision, and appointed Omar Sultan Al Olama as the world’s first Minister of State for Artificial Intelligence. The strategy set a goal of AI leadership by 2031 across sectors including transport, health, education, energy and government services — explicitly framed around reducing long-term dependence on oil revenue through new, technology-led industries.
What Is Stargate UAE?
A 1-GW cluster inside the 5-GW campus — not a synonym for it
Stargate UAE is a 1-gigawatt AI compute cluster in Abu Dhabi, built by G42 (through its Khazna Data Centers subsidiary) and operated by OpenAI and Oracle, launched on 22 May 2025 with Nvidia, Cisco and SoftBank as partners. It sits inside the larger 5-GW UAE-U.S. AI Campus announced days earlier, and is the first named, publicly tracked construction project within it. As of this update, only its first 200-megawatt phase had a public completion target — Q3 2026.
Treat any headline that uses “Stargate UAE” and “the 5-GW campus” interchangeably with caution. Stargate UAE is real, named, and under construction — but it is roughly a fifth of the campus’s ultimate power target, and the campus itself is still mostly a plan.
Who Is Building the UAE’s AI Infrastructure?
Entity map: government, company, investor and technology-partner roles, kept separate
| Entity | Type | Role |
|---|---|---|
| UAE Government | National policy | Set the 2017 AI strategy; negotiated the May 2025 U.S.-UAE AI Acceleration Partnership that underpins the campus and chip-access rules. |
| G42 | AI technology group | Abu Dhabi-based holding company, founded 2018; leads construction of the 5-GW campus and Stargate UAE through subsidiary Khazna Data Centers. |
| MGX | Investment fund | Abu Dhabi state-backed AI investment vehicle, founded by Mubadala and G42 in March 2025; a minority investor in Khazna Data Centers. |
| Khazna Data Centers | Data-center operator | G42’s data-center infrastructure subsidiary and majority-owned by G42; MGX and Silver Lake hold minority stakes after e& sold its 40% stake for $2.2bn. |
| Microsoft | Strategic investor/partner | Invested $1.5bn in G42 in April 2024; Brad Smith sits on G42’s board; G42 runs on Microsoft Azure under a governed security agreement. |
| OpenAI | Compute operator | Co-operates the Stargate UAE compute cluster alongside Oracle; a named partner in the May 2025 launch, not the campus’s sole builder. |
| Stargate UAE | Named project | The 1-GW compute cluster inside the 5-GW campus — the entity most often confused with the campus itself. |
| Nvidia | Chip supplier | Supplies AI accelerators (including Blackwell-class GPUs) subject to U.S. export approvals; also a named Stargate UAE partner. |
| Oracle, Cisco, SoftBank | Technology partners | Named partners in the Stargate UAE consortium alongside G42, OpenAI and Nvidia; roles span cloud infrastructure, networking and investment. |
What Does an AI Data Center Actually Need?
A physical industrial system, not just software in a building
Every one of these layers is a physical constraint. A model can be copied in seconds; a power station, a cooling plant or a fiber trench cannot. That mismatch — instant software, slow infrastructure — is why building AI capacity at gigawatt scale takes years, not product-launch cycles.
Why Electricity May Matter as Much as AI Chips
A multi-gigawatt campus is also an energy-infrastructure project
AI accelerators draw power continuously under heavy load, and the servers around them generate proportional heat, which then needs its own energy-hungry cooling. Grid connections and backup power have to be sized for the whole system, not just the chips. That is why a “5-GW AI campus” is, in practical terms, also a power-plant and grid-connection project.
G42’s own announcement for the 5-GW UAE-U.S. AI Campus states it will draw on a mix of nuclear, solar and natural gas power once complete, aiming to limit emissions at that scale. That is the company’s stated design goal for the finished site — not a documented, audited energy mix for the portion currently under construction. No verified public breakdown of exactly how much of the built 200-MW phase runs on which power source exists as of this update, so this article does not assign a specific energy-mix percentage to it.
Power Sources Named for the Finished Campus
- Natural gas — the UAE’s dominant existing power-generation fuel, and the most readily available baseload source for near-term capacity.
- Solar — the UAE has invested heavily in utility-scale solar (including Masdar-linked projects) as part of its broader energy diversification.
- Nuclear — the UAE already operates the Barakah nuclear plant, the Arab world’s first, giving it an existing large-scale, low-carbon baseload option to draw on.
How Do You Cool an AI Data Center in the Desert?
Ambient heat and water access are real engineering constraints here, not incidental details
Cooling options range from simple to complex: air cooling is the cheapest but least effective for dense AI racks; chilled-water systems handle more heat per rack but require a stable water supply; direct-to-chip liquid cooling pipes coolant straight to the hottest components and is increasingly standard for the newest GPU generations; immersion cooling, submerging hardware in dielectric fluid, is used in some high-density deployments but remains less common at hyperscale.
Two efficiency metrics are widely used to describe a data center’s overhead: PUE (Power Usage Effectiveness) compares total facility energy use to the energy actually used by computing equipment — a PUE of 1.5 means 50% more power is spent running the building than running its computers. WUE (Water Usage Effectiveness) measures water consumed per unit of computing energy. Ambient desert heat pushes cooling systems harder than a temperate-climate site would need, which is one reason liquid and hybrid cooling are getting more attention in Gulf data-center planning generally.
No official, facility-specific water-consumption figure has been published for the UAE’s 5-GW campus or Stargate UAE’s built phase. Generic global data-center water-use averages exist in industry literature, but applying one directly to this specific project without a documented source would be a guess, not a fact — so this article does not quote one.
Why AI Chips Became a Geopolitical Issue
Advanced accelerators now move through diplomacy, not just procurement
The most advanced AI accelerators — chips like Nvidia’s Blackwell-class GPUs — are subject to U.S. export controls because of their potential military and strategic applications. For years, the UAE sat in a more restrictive export tier, requiring individual licenses and raising U.S. concern over G42’s historical technology ties to China.
That changed through 2024–2026: Microsoft’s 2024 investment in G42 came bundled with an intergovernmental security-assurance agreement; the May 2025 U.S.-UAE AI Acceleration Partnership created a government-to-government framework for “trusted” AI infrastructure; specific chip export approvals for G42 followed in November 2025; and in July 2026, the U.S. Commerce Department formally upgraded the UAE to its top export-control tier (Country Group A:5), enabling license-free imports of advanced chips for approved entities.
The throughline: GPUs at this scale are no longer treated as ordinary commercial hardware. They carry national-security conditions, reporting requirements and government-to-government trust frameworks — which is exactly why a data-center construction story and a diplomatic story became the same story.
Interactive: Mega Campus vs Distributed AI
Neither approach is automatically safer or cheaper — the trade-offs are real on both sides
Which Would You Build?
There is no universally “correct” answer here — a single campus is more efficient to build and run; distributed infrastructure is more resilient to any one localized disruption. Reported 2026 UAE planning appears to be trading some of the former for more of the latter, not choosing one extreme.
Why 2026 Changed the Data-Center Debate
What was reported, what was confirmed, and what planners are now said to be considering — kept separate
What was reported: Around 1 March 2026, drones or missiles reportedly struck three AWS data centers — two in the UAE, one in Bahrain — during the wider Iran conflict, forcing them offline. Reported disruptions extended to banking, payments, delivery apps and enterprise software across the region.
What was confirmed: AWS declined to comment on the strikes. Iran’s Fars News Agency claimed the facilities supported military and intelligence activity — a claim from a party to the conflict, not an independently verified fact. Neither UAE nor Bahraini officials are on record confirming operational details.
What planners are reportedly considering: On 11 September 2026, Reuters reported that UAE officials, per people familiar with internal discussions, are now weighing whether to distribute the 5-GW campus’s capacity across multiple sites, and whether to build some facilities underground or otherwise harden them against drone and missile attack, alongside additional air-defense coverage. This is reported deliberation, not an announced final decision.
The evidence-based framing: the disruption highlighted the risks of concentrating critical computing, energy and communications infrastructure geographically — not that “the war proved the project would fail.” The 5-GW campus has not been cancelled, and Stargate UAE’s construction timeline had not been publicly revised as of this update.
How Do You Protect “The Cloud”?
Infrastructure-resilience concepts, at a level useful to a general reader
Resilience Concepts Actually Used in This Industry
- Geographic redundancy — running the same workload from more than one physical location, so no single site is a single point of failure.
- Multiple availability zones — a cloud-industry standard practice of spreading capacity across distinct facilities within or across regions.
- Backup power and diverse network paths — generators, batteries and more than one fiber route in and out of a site.
- Workload replication and disaster recovery — keeping copies of critical systems and data ready to fail over quickly.
- Physical hardening — construction choices (including underground siting, in some industries) that reduce vulnerability to blast, fire or structural damage.
- Cybersecurity and supply-chain resilience — protecting the software and hardware supply chain feeding the facility, not just its walls.
You can move software in seconds. You cannot move a power station, a cooling plant or a fiber trench nearly as fast — which is exactly the gap 2026 exposed.
Interactive: Where Should the AI Live?
Hypothetical siting concepts — not a map of any real, sensitive facility
🏙️ Urban / Industrial Area
Connectivity: High. Land: Limited. Infrastructure access: High. Concentration risk: Potentially higher, given denser surrounding population and infrastructure.
🏜️ Desert Site
Land availability: High. Scale potential: High — this is the model behind the actual announced 10-sq-mi campus. Cooling challenge: Significant, given ambient heat and water access questions.
⛰️ Protected-Terrain Concept
Physical protection potential: Higher than an open desert site. Construction complexity: High. This is a concept discussed in 2026 reporting as an option under consideration — not a confirmed UAE project.
👋 Underground Concept
Protection potential: Higher. Construction: Complex and costly. Cooling and maintenance: Both significantly more complex than a surface facility. Reuters reported this as an option UAE planners are weighing, not an approved or built facility.
| Concept | Security | Cost | Cooling | Connectivity | Power Access |
|---|---|---|---|---|---|
| Urban/Industrial | Lower | Lower | Moderate | Highest | High |
| Desert | Moderate | Moderate | Hardest | Moderate | Highest (space for on-site generation) |
| Protected Terrain | Higher | High | Moderate–Hard | Lower | Moderate |
| Underground | Highest | Highest | Hardest | Lower | Moderate |
Every ranking here is a directional engineering trade-off, not an official UAE assessment. No underground or mountain-based AI facility has been confirmed as an approved UAE project as of this update — only reported as one option among several under discussion.
Oil Field vs AI Data Center
The Gulf’s list of things worth protecting is expanding, not being replaced
20th-Century vs AI-Era Strategic Infrastructure
The Gulf’s critical-infrastructure mix is expanding, not swapping one asset for another. Hydrocarbons remain enormously important to the UAE’s economy and export revenue. What’s changed is that computing infrastructure — chips, power and cooling plants, fiber routes — is becoming an additional strategic asset alongside oil and gas, as the country’s economic diversification continues.
UAE Economic Diversification and AI
Technology-led diversification, not a “post-oil” story
The UAE’s AI push traces directly back to the 2017 national strategy’s stated goal of reducing long-term dependence on oil revenue through new, technology-led industries. Nearly a decade later, AI infrastructure investment — MBZUAI, G42, the 5-GW campus, MGX’s global data-center financing — is one of the clearest expressions of that goal in practice.
The September 2026 UAE-Germany €40 billion pledge fits the same broader pattern: energy and technology cooperation (LNG, renewables, hydrogen, AI, digital infrastructure) as part of expanding the UAE’s economic base beyond hydrocarbons, not evidence the country is abandoning oil and gas — ADNOC, the UAE’s national oil company, was itself one of the parties signing new energy deals in Germany that same week.
What Happens Next?
Open questions, not predictions presented as fact
Watch These, Without Assuming the Outcome
- Distributed AI infrastructure — whether the UAE formally adopts a multi-site design for some or all of the remaining 5-GW capacity.
- Hardened or underground facilities — whether any of the reported protective-construction options move from consideration to an approved, funded project.
- Greater redundancy and independent power — how much on-site generation (gas, solar, eventually nuclear) the campus adds as it scales past its first phase.
- Advanced cooling adoption — whether liquid or hybrid cooling becomes the default as later, denser phases are built.
- Secure fiber routing — whether additional redundant network paths are added alongside any new sites.
Explore More Timelines
People Also Ask
Common related questions, answered independently
Frequently Asked Questions
Independent, self-contained answers