HUMAIN and NVIDIA plan 500 megawatts of AI factories in Saudi Arabia

HUMAIN and NVIDIA announced a new strategic partnership at LEAP 2026 to build what the two companies are calling AI factories in Saudi Arabia. HUMAIN is making what the companies describe as a major investment toward a projected capacity of up to 500 megawatts, powered by several hundred thousand of NVIDIA's most advanced GPUs, over the next five years. "AI factory" is NVIDIA's own term for the category, and it is worth taking literally rather than as marketing gloss. The company uses it to distinguish facilities purpose-built to turn electricity and silicon into trained and served AI models at industrial scale from general-purpose cloud data centres that happen to run some AI workloads alongside everything else. The distinction matters for how these sites get designed: denser racks, heavier power and cooling per square metre, and networking built around keeping thousands of GPUs working as a single machine rather than thousands of independent ones. The first phase of the build, according to NVIDIA, will be an 18,000-GPU system based on the NVIDIA GB300 Grace Blackwell platform, connected with NVIDIA's InfiniBand networking. Grace Blackwell pairs NVIDIA's own Grace CPU with its Blackwell-generation GPUs on the same module, a design aimed at feeding data to the GPUs fast enough that they are not left waiting. InfiniBand is the high-speed interconnect fabric that links GPUs to each other; at this scale, the quality of that fabric determines whether 18,000 GPUs can actually train a single frontier-sized model together, or merely sit in the same building. The stated purpose, per NVIDIA's announcement, is to provide secure foundational infrastructure for training and deploying sovereign AI models at scale, serving industries in Saudi Arabia and, the company says, beyond. Saudi officials framed NVIDIA's plan as one of the standout announcements among the more than $15 billion in technology investments unveiled as LEAP 2026 opened. It is important to separate this from an earlier, similarly-worded NVIDIA announcement. In November 2025, at the US-Saudi Investment Forum, HUMAIN expanded its partnership with NVIDIA alongside xAI, Global AI and AWS — the announcement that included xAI's own 500-megawatt Saudi data centre. Coverage of that expansion has recirculated during LEAP week, but it is not this. What NVIDIA and HUMAIN announced at LEAP 2026 is a separate, bilateral AI-factory commitment between the two companies, with its own capacity target and its own first-phase hardware specification. It also fits a pattern rather than standing alone. HUMAIN has spent the past year assembling compute capacity from multiple vendors rather than betting on one architecture: it has AMD Instinct GPU systems already live with Cisco , a 250-megawatt data centre project with Together AI , and a 100-megawatt first phase of a larger DataVolt campus at NEOM . NVIDIA's 500 megawatts is now the largest single figure in that stack, but it sits alongside rather than replaces the others. What the announcement does not include is as notable as what it does. There is no disclosed dollar value for HUMAIN's investment, no exact GPU count behind "several hundred thousand," and no delivery date for the 18,000-GPU first phase. The 500 megawatts is a projected capacity over a five-year window, not a built or contracted amount, and nothing in the release specifies where in the Kingdom the factories will sit. TechScoop is reporting from LEAP 2026 in Riyadh.