Export approval is the gating item for Saudi AI compute
Category: AI & Data
Published: 2026-09-03T05:30:00.000Z
HUMAIN has local approval to buy 18,000 Nvidia AI chips, with the purchase subject to US export clearance. Construction schedules and licensing schedules are independent variables.
HUMAIN, the PIF-owned Saudi AI company, has received local regulatory approval to purchase 18,000 of Nvidia's latest AI accelerators for its first data centres. The purchase remains contingent on governance protocols and on approval from the United States government. Every element of the Saudi AI programme that can be decided in Riyadh has been decided. The remaining variable sits with an external licensing process. HUMAIN broke ground on facilities in Riyadh and Dammam, each with initial capacity of up to 100MW, expected live in the second quarter of 2026. The roadmap runs to 1.9GW by 2030 and six gigawatts by 2034. Concrete, power and cooling can be delivered on that schedule with capital and competent contracting. Accelerators run on a different timetable, because supply is allocated by a manufacturer with more demand than output, under an export regime that treats advanced compute as a strategic control point. The practical consequence is that construction and licensing are independent variables, and both have to land for capacity to come online. That is why Gulf AI programmes have consistently been structured around partnerships with American companies rather than around procurement alone. Stargate UAE — a one-gigawatt compute cluster inside a planned five-gigawatt AI campus at Masdar City, built by Khazna Data Centers with OpenAI, Oracle, Nvidia, Cisco and SoftBank — is one version of that structure. Rather than buying chips and awaiting a licence, the UAE assembled a consortium in which American companies are principals with their own commercial interest in the licence being granted. HUMAIN's reported joint venture with AMD, under which AMD would take equity in a Saudi special purpose vehicle, follows the same logic. Giving the supplier equity turns a vendor relationship into a shared interest. The second-order effect reaches the region's startups. For a company in Riyadh or Cairo trying to train a model, the practical question is whether accelerator time can be bought at a competitive price in the region at all. Today most regional AI companies build on top of hosted foreign models rather than training their own, which is a function of where compute is available and at what price. Domestic compute at domestic prices, under domestic data residency rules, is what would let a Saudi company train at scale locally — and that is precisely what HUMAIN's capacity is being built to provide. Until it arrives, the region's AI sector operates largely as an applications layer on infrastructure hosted elsewhere, which describes most of the world's AI sector today. HUMAIN's stated ambition is for Saudi Arabia to be the world's third-largest AI provider. Reaching it depends on sustained access to advanced silicon over a decade, which makes the supply relationship as central to the programme as the construction schedule — and it is the part the joint venture structures are designed to secure.