Inside the Race to Build “Sovereign AI” in the Gulf

When finished, the data center campus in the desert outside of Abu Dhabi is expected to be the world’s largest AI infrastructure site. The scale is hard for humans to comprehend: buildings built to house GPU clusters that use gigawatts of power, cooling systems designed for a climate where temperatures frequently rise above 40 degrees Celsius, and fiber networks connecting facilities that will eventually house models trained on data that never crosses national borders. Separate from any geopolitical framing, the physical aspirations of what is being constructed there make a message of its own: the Gulf states have determined that the AI era is not something to passively consume, and they are investing appropriately.

This development is motivated by the idea of “sovereign AI,” which holds that a country’s AI infrastructure, models, data, and hardware should all be controlled domestically rather than on servers held by Chinese or American companies. The most ardent supporters of this framework have been Saudi Arabia and the United Arab Emirates, and since 2025, investments in it have advanced the discussion from ambition to actual reality at a rate that the technology policy world has been trying to keep up with. HUMAIN, a specialized AI firm founded by Saudi Arabia’s Public Investment Fund, has been using 18,000 Nvidia Blackwell processors in AI factories built to function without constant human interaction. The collaboration with Groq and Nvidia points to a procurement strategy that maintains Saudi operational control over the infrastructure while hedging across hardware manufacturers.

The UAE’s most prominent sovereign AI vehicle has been Abu Dhabi’s G42, which has developed supercomputing capacity in collaboration with Cerebras and released K2 Think, an open-source model with 32 billion parameters that exemplifies a particular kind of capability statement: the Gulf can create foundation models, not just use foreign ones. The UAE appears to be attempting to acquire portions of the supply chain rather than merely acquiring access to its outputs, as seen by the capital commitments made by MGX, a more recently established sovereign investment arm, to chip production and international foundational model alliances.

The Falcon model, created by the Technology Innovation Institute in Abu Dhabi, has been accessible as an open-source release for a while and has garnered genuine international usage, demonstrating that the caliber of models developed in the Gulf has surpassed a threshold that makes them competitive beyond regional applications.

The sovereign AI aim most closely relates to a particular market failure in global AI development in the Arabic language dimension. With Arabic making up a minor portion of the corpus and frequently lacking the cultural and dialectical complexity that 400 million Arabic speakers would identify as true comprehension, the main American language models were mostly constructed on English-language training data.

A model developed to comprehend the unique registers and contexts of Arabic as it is actually used in the Middle East and North Africa and trained on data from the Gulf region fills a gap in the market. HUMAIN is integrating Saudi Arabia’s huge language model, ALLaM, which was created especially to fill this need. The market it addresses is real and underserved, but it has to be seen if it bridges that gap at the level of quality necessary for commercial and government adoption.

The “vassal AI” concept, which is being used by Gulf technology authorities to characterize the alternative scenario in which countries become reliant on foreign companies for essential technology infrastructure, accurately conveys the strategic fear that motivates this investment.

Gulf leaders explicitly draw a comparison to energy dependence: they are not inclined to replicate that dependency structure in artificial intelligence because they built modern economies on a resource controlled by geography and then witnessed the geopolitical leverage that resource concentration creates. From this angle, the data center capacity being constructed in Riyadh and Abu Dhabi is comparable to the infrastructure for oil storage and processing that initially gave the Gulf its economic clout. The resource is not the same. The reasoning is the same.

Observing the capital deployment and the ongoing building gives the impression that the Gulf’s independent AI drive is further advanced than most Western technology commentary has taken into consideration. The region lacks the density of AI researchers and data scientists that Silicon Valley, London, or Toronto have amassed, and the U.S. export restrictions on cutting-edge semiconductors provide continuous procurement complexity that no amount of money can completely fix.

However, the energy advantage is tangible: the Gulf offers both hydrocarbon generation and growing solar capacity at a cost structure that most other locations cannot match, and the computing infrastructure these data centers require consumes massive amounts of power. Within this decade, the Gulf may have a sovereign AI infrastructure that functions at true scale thanks to a mix of wealth, energy, and political will. It’s just as likely that closing the skill gap will be more difficult than closing the hardware deficit. In any case, construction is still ongoing.