Stargate UAE switches on its first 200 MW.

In Abu Dhabi, a new AI campus is about to draw its first 200 megawatts. In Riyadh, a rival build is scaling in parallel. The Gulf is not choosing between energy and compute. It is building both at once, from mixed sources, on purpose.

Stargate UAE brings its first 200 megawatt (MW) phase online in September 2026. It is the opening slice of a 5 gigawatt UAE-US AI campus in Abu Dhabi. The Emirati group G42 leads the build. The site runs on a tri-fuel mix: nuclear from Barakah, utility-scale solar, and natural gas. A gigawatt (GW) is one thousand megawatts.

The template is simple. Build the power and the compute together.

The headline is the hardware. The real move is the power behind it. The first 200 MW is due to complete in 2026, with the full campus spanning roughly 10 square miles. It draws dedicated nuclear baseload, solar by day, and gas for backup. No single source carries the load alone. Stargate UAE itself debuts as a 1 GW cluster within that campus. Oracle, Nvidia, SoftBank and Cisco supply the infrastructure and chips. That mix is what lets construction and generation advance side by side. The compute arrives because the power arrives with it.

The UAE did not start from nothing. Its Barakah plant now runs four reactors at 5.6 GW of capacity, the fourth unit online since March 2024. The station generates about 40 terawatt-hours a year, near a quarter of national electricity. Much of that output already serves existing demand. So new AI campuses add dedicated and mixed supply rather than lean on the grid. Unit 1 first reached the grid in August 2020. Barakah is the first nuclear plant in the Arab world. It avoids about 22.4 million tonnes of carbon dioxide a year. The baseload is proven, and the model is now repeatable.

The pattern crosses the border. Saudi Arabia's HUMAIN, backed by the Public Investment Fund (PIF), is assembling compute at scale. It plans to grow from 1.9 gigawatts by 2030 to 6.6 gigawatts by 2034. The build is worth about £58bn ($77bn, €68bn) at market rates. Its first parks, twin campuses of up to 100 MW in Riyadh and Dammam, go live in 2026. The power mix is gas and solar, with no operating nuclear yet. The template holds, even with a different fuel base.

Here is the core idea. The Gulf treats generation and compute as one project. Sovereign capital funds both, so neither waits on the other. There is no queue for a shared grid connection. Dedicated supply is designed in from the first drawing. Land, water and permits are arranged as one package. The developer controls the whole chain, from fuel to rack. Construction of the plant and the data hall runs in parallel. The result is speed that a grid-dependent build cannot match.

The tri-fuel choice is deliberate engineering, not indecision. Nuclear gives firm, low-carbon baseload around the clock. Solar adds cheap daytime energy in a sun-rich region. Gas covers the ramps and the gaps. Each source answers the others' weaknesses. Together they hold the steady 24/7 supply that AI hardware demands. AI loads swing fast and hard. A blended supply absorbs those swings better than one fuel. Resilience and speed come from the blend, not from any one fuel.

Other regions solve the same problem differently. In the United States, hyperscalers buy their way to power. They sign power purchase agreements (PPAs) and build gas on site. ExxonMobil and NextEra are developing a 1.2 GW gas plant with carbon capture for data centre demand. The capital is private, not sovereign. The tech sector signed about 40% of corporate renewable power deals in 2025. Its on-site gas projects, though, remain early-stage. The approach is fast, but it stays exposed to grid queues and turbine backlogs.

Asia offers a third model. South Korea's AI data centre (AIDC) Alliance, launched in July 2026, targets 18.4 GW of capacity by 2035. Planned investment runs to about £487bn ($648bn, €570bn), with the state organising private capital at scale. NAVER, with Nvidia and Brookfield, is expanding one AI data centre to 200 MW. The stated path runs to a gigawatt. Seoul treats sovereign compute as national industry. A single gigawatt of AI compute needs roughly 450,000 top graphics processing units (GPUs). It draws about 8.76 terawatt-hours a year. The financing is state-led, and the ambition is explicit.

The contrast with Europe is stark. France is committed to six new reactors, at a programme cost of £62bn ($83bn, €73bn). The first is not expected before 2038. Across Europe, grid connection queues stall comparable campuses for years. Planning approval and grid access move on separate, slow tracks. Neither waits for the other, and both take years. Where energy and compute are planned apart, delivery slips. The Gulf lesson is that separation is the slow path.

The capital case is about time, not just cost. Co-locating generation and compute removes the interconnection wait. That wait, not the capital, is often the scarcest asset. Stargate UAE's campus reflects investment of more than £23bn ($30bn, €26bn). Sovereign balance sheets and mixed sources bring campuses online in quarters. A grid-dependent build measures the same step in years. Every quarter saved brings compute online earning revenue sooner. That saved time is the real return the Gulf model captures. Underwriting that ignores the energy schedule misprices the whole asset.

The Gulf switch-on is a template, not a one-off. Energy and compute, built together, from mixed sources, on sovereign capital. From Abu Dhabi to Riyadh to Sejong, the winners pair the megawatt with the model. Intelligence can map which sites clear first. The template will spread wherever sovereign capital meets abundant, mixed fuel. The chips still need the power beneath them.

Next week: Nvidia calls AI a power-limited industry. The interconnection queue proves it.

"When I think of revenue, I think of Attio." - Shreman Shrestha, Head of Business at Granola

Here's what that adds up to:

Start Your Trial