When people talk about "AI in the Gulf", they often mean something that sounds abstract: agreements, gigawatts, billion-dollar funds. This page tries to make it concrete. Modern AI rests on four foundations: computers that train and run models, electricity to power them, the models themselves, and the people who build and look after all of it. For each one we say what exists today, what has only been announced, and what is still uncertain.

A few terms, in plain words

  • Compute: the processing power used to train and run AI. It mostly comes from specialised chips called GPUs (graphics processing units), housed in large buildings called data centres.
  • Megawatt (MW) / gigawatt (GW): units of electrical power. One gigawatt is 1,000 megawatts. When a data centre is described as "1GW", that refers to the power it is designed to draw, not to how "smart" it is.
  • Model / LLM: a large language model is software trained on huge amounts of text so that it can answer questions, summarise and write. "Parameters" are the internal settings learned during training, and they are a rough measure of size.
  • Open weights / open source: the trained model is published so that others can download, inspect and build on it.

1. Compute: big plans, and a careful distinction

Most of the region's headline numbers are about data centres, so it helps to read them slowly.

In the UAE, the US Embassy described the 5GW UAE–US AI Campus, unveiled in Abu Dhabi on 15 May 2025, as the largest of its kind outside the United States. It was planned across 10 square miles and powered by a mix of nuclear, solar and gas. The same release says the compute would be reserved for US hyperscalers and approved cloud providers, under "Know-Your-Customer" rules meant to stop the technology being diverted to others (US Embassy UAE). Within that campus, OpenAI announced Stargate UAE on 22 May 2025: a 1GW cluster, with a first 200MW phase that OpenAI said was expected to go live in 2026 (OpenAI).

Two honest caveats belong next to those numbers. First, a gigawatt plan is a design target, not a working facility. As of 28 September 2026 we have found no official confirmation that the first 200MW is in service. Second, the plan itself has changed. After drone strikes on data centres in March 2026, Reuters reported (11 September 2026, citing four sources) that the campus is being redesigned as a network of dispersed, more protected sites. G42 says the work remains on track (CNBC summary). We cover this in more depth in Resilience in 2026.

In Saudi Arabia, the PIF-owned company HUMAIN announced on 19 November 2025 that it planned to deploy up to 600,000 NVIDIA GPUs over three years, in the Kingdom and in the US. It also announced a data-centre network with xAI anchored by a facility of 500MW or more, and an "AI Zone" in Riyadh with AWS of up to 150,000 GPUs (HUMAIN release). The release carries a standard "forward-looking statements" caveat. These are plans, and they are reported as such.

Chips need permission. Advanced AI chips are export-controlled by the United States. In November 2025 the US Commerce Department authorised G42 and HUMAIN to buy the equivalent of up to 35,000 NVIDIA GB300 chips each, subject to security and reporting requirements (The National). This shapes the whole regional story. The Gulf's compute build-out takes place inside a close security relationship with the US, not outside it.

Why this matters for people, not just engineers: local compute means lower delays for services used in the region. It also lets governments and hospitals keep sensitive data in the country, and gives local researchers access to hardware that used to be available only abroad. OpenAI, for example, says a UAE cluster could serve people within a 2,000-mile radius, "reaching up to half the world's population". That is a company's own claim about reach, not a measure of use.

2. Electricity: the region's quiet advantage

AI uses a lot of electricity, but it helps to keep that in proportion. The International Energy Agency (IEA) estimates that data centres used about 1.5% of the world's electricity in 2024 (around 415 TWh), and projects roughly 945 TWh by 2030. It also finds that data centres account for about a tenth of global electricity-demand growth to 2030, which is less than air conditioning, electric vehicles or industrial motors (IEA, Energy and AI, 10 Apr 2025). Its overall judgement is measured: "Concerns that AI could accelerate climate change appear overstated, as do expectations that AI alone will address the issue."

The Gulf starts from an unusual position. It has abundant land, strong solar resources, established gas infrastructure and, in the UAE, nuclear power. The Barakah nuclear plant reached full four-unit operation on 5 September 2024. Its operator, ENEC, says it generates about 40 TWh a year, "up to 25% of the UAE's electricity needs" (ENEC). A steady, low-carbon supply matters for data centres, which need power around the clock.

The IEA also notes something that rarely makes headlines: AI can make grids themselves more reliable. It estimates that AI-based fault detection can cut outage durations by 30–50%. For a region building new cities and grids quickly, that is a practical benefit, not a speculative one.

3. Models: Arabic, open and increasingly capable

For many years, the most capable AI models were built almost entirely in the US and China, and trained mostly on English and Chinese text. The Gulf now contributes models of its own, and several are openly published.

  • Falcon (UAE, Technology Innovation Institute). On 5 January 2026 TII released Falcon H1R 7B, a small "reasoning" model. TII says it outperforms several larger open models on key benchmarks, and scores 88.1% on the AIME-24 maths test. It is released under the Falcon TII licence (TII). Those rankings are TII's claims, and independent evaluations should be checked before repeating them.
  • K2 Think and Jais (UAE, MBZUAI and G42). K2 Think, launched on 9 September 2025, is a 32-billion-parameter reasoning system. Its builders released the training data, weights and code, which is unusually open (G42). Jais, first released in 2023, is an Arabic-centric model from the same ecosystem.
  • ALLaM and HUMAIN Chat (Saudi Arabia). HUMAIN Chat launched in August 2025, powered by ALLaM 34B, an Arabic-first bilingual model that grew out of SDAIA's National Center for AI.
  • Fanar (Qatar). Inaugurated on 10 December 2024 by Prime Minister Sheikh Mohammed bin Abdulrahman Al Thani, Fanar was developed by QCRI at Hamad Bin Khalifa University. HBKU says it drew on more than 300 billion words, with contributions from the Ministry of Endowments, Qatar National Library, Al Jazeera and others (HBKU).

Why Arabic-first models matter. Research presented at ACL 2024 found that many language models, including some trained specifically on Arabic, tended to favour Western names, foods and places even when prompted in Arabic, and sometimes stereotyped Arab entities (Naous et al., ACL 2024). Models trained on carefully chosen Arabic sources, with regional institutions involved, are one practical response. They are not a guarantee of fairness, but they are a step toward AI that understands the people who use it.

4. People: the foundation that lasts longest

Buildings and chips age quickly. Skills stay.

  • Universities. MBZUAI in Abu Dhabi, founded in 2019 according to the US Embassy release, took its first students in January 2021 and describes itself as the first graduate-level, research-based AI university.
  • Schools. From the 2025–26 academic year, AI became a formal part of the UAE public-school curriculum from kindergarten to Grade 12. About a quarter of the content is on ethics (Gulf News).
  • Skills at scale. Microsoft, as part of a $15.2bn UAE investment between 2023 and 2029, has committed to help train one million people in the UAE by the end of 2027 (Microsoft).
  • Talent flows. Stanford's AI Index 2026, as summarised by a regional outlet, reports that the UAE saw the largest increase in AI talent concentration of any country between 2019 and 2025, and that Saudi Arabia ranked fourth for net AI talent migration (Middle East AI News; to be confirmed against the Stanford report itself).

What we don't know yet

  • Whether announced compute (hundreds of thousands of GPUs, multiple gigawatts) will be delivered on the stated timelines, especially under wartime pressure.
  • How much of that compute will serve people in the region, and how much will serve global customers.
  • Whether open Arabic models will be widely adopted, or remain mainly research showcases.
  • The long-term water and energy footprint of new facilities in a hot climate. Operators should publish this data.

What this means for you

If you live or work in the Gulf, the practical effects are gradual rather than dramatic. Services should get faster. More of your data may stay in the country. More courses, jobs and research roles will be available close to home. You don't need to follow gigawatt announcements to benefit from them. It is still reasonable to ask of every announcement: is it built, is it running, and who does it serve?

Sources are listed inline. Every figure on this page is recorded in the platform's fact base with its verification level. Claims and projections are tracked in the claims register.