Three signals I watch in Gulf AI
Compute build-out, local content rules and talent programs: the verified facts behind what I track in Saudi AI.
By Ali Asif,
People ask me which Gulf AI headlines matter. I ignore most of them and track three things that tell me how the market is being built. Each one has a primary source behind it.
Signal one: sovereign compute
In May 2025, NVIDIA announced that HUMAIN, a PIF company focused on AI, would build AI factories in the Kingdom with up to 500 megawatts of capacity, powered by several hundred thousand NVIDIA GPUs over five years. The first phase is an 18,000-chip GB300 supercomputer. The same release says SDAIA, the national data and AI authority, will deploy up to 5,000 Blackwell GPUs for a sovereign AI factory.
DatacenterDynamics reported that HUMAIN broke ground on two data centers, in Riyadh and Dammam, with an initial 100 MW each.
One caution on reading these numbers. A groundbreaking and a chip order are plans in motion, and the useful evidence arrives later, when a site is energised and a customer runs real work on it. I treat the announced figures as direction, not as delivered capacity.
Why it matters to me: compute decides which AI products can be built and hosted inside the Kingdom. For my own work that affects whether a client with sensitive tender documents can use a model without data leaving the country. I'm watching for named customers and live workloads on this capacity, since those show demand more clearly than announced megawatts.
Signal two: local content rules
The Kingdom has a clear policy that national products and providers get a lift in government buying. Saudipedia's summary describes a 10% price preference for national products, a Mandatory List maintained by the local content authority, and, on certain contracts, a financial evaluation that weights local content at 40% against price at 60%.
For AI this shapes who sells what. A foreign model vendor that partners with a Saudi company and builds local capability is in a stronger position when the buyer scores local content. I read it as a straightforward invitation: build here, hire here, and the scoring reflects it. Founders who plan for that from day one move faster than those who treat it as a later compliance step.
Signal three: talent programs
SDAIA, together with the education and human resources authorities, announced in late 2025 that its SAMAI initiative had trained and certified more than one million citizens in AI. The education authority's release reports that 52% of participants were women, 70% were employed and 30% were students.
That is a literacy number, not an engineering number, and I read it as such. But literacy at that scale changes the buyer. Procurement managers, engineers and finance teams who've completed a basic course ask better questions of a vendor. For anyone selling AI services here, the sales conversation gets shorter when the other side already knows the terms.
How the three connect
Compute gives the Kingdom somewhere to run models. Local content rules create a reason to build and hire locally. Talent programs create buyers and staff who can use the result. If I saw one move without the others, I'd read it as a headline. When all three move together, I read it as a market forming.
What I'd watch next
Customer announcements on the new compute. Updates to the Mandatory List that touch software and AI services. Follow-on programs after SAMAI that go beyond awareness into applied skills.
I'll update this note when any of those lands. Dated sources only, and I'll fix anything I got wrong.