Saudi Arabia's Humain expands Nvidia deal to 84MW of AI compute

Summary

• Humain and Nvidia will build up to 84MW of AI computing capacity inside Saudi Arabia.
• The two-phase plan covers roughly 27,000 GPUs, starting with Blackwell systems in late 2026.
• Regional buyers gain domestic access to Nvidia's newest architectures instead of renting capacity abroad.

The latest

Up to 84 megawatts of artificial intelligence computing capacity — about 27,000 graphics processing units — will be built in Saudi Arabia under an expanded partnership between state-backed Humain and Nvidia, the two companies said. The agreement was announced on the sidelines of Nvidia's GTC conference in Berlin on Thursday, 22 October 2026, and runs in two phases through mid-2027.

Details

• Phase one: The first tranche covers 14MW of Nvidia Blackwell HGX B300 systems, roughly 6,000 GPUs, serving Humain's cloud customers across AI companies, government entities and enterprises. Deployment is scheduled to begin in the fourth quarter of 2026, making it the nearer-term and more firmly scheduled half of the plan.
• Phase two: The larger tranche is 70MW built on Nvidia's next-generation Vera Rubin architecture, about 21,000 GPUs. It splits into 50MW for an unnamed American customer and 20MW for Humain's own cloud, with deployment starting in June 2027 for the US client and July or August 2027 for Humain.
• The caveat: Both companies said the Vera Rubin projects remain subject to definitive agreements and deployment plans, meaning the 70MW portion is not yet locked in. The two sides did not disclose the value of the deal or the identity of the American customer taking the larger share.
• The CEO: Tareq Amin, chief executive of Humain, told The National the company is "moving quickly to bring the next generations of Nvidia-accelerated computing into Saudi Arabia at meaningful scale." Humain is owned by the Public Investment Fund and chaired by Crown Prince Mohammed bin Salman.
• The origin: The expansion builds on a May 2025 agreement announced during US President Donald Trump's visit to Riyadh, when Nvidia said it would supply Humain with hundreds of thousands of advanced chips over five years. That deal's first phase involved 18,000 Blackwell processors within a plan reaching 500MW over five years.
• The ambition: Humain is planning around 3 gigawatts of data centre capacity by 2030 and more than 6 gigawatts by 2034, a scale that places the current 84MW announcement as an early increment rather than the endpoint of the build-out.
• The rivalry: Saudi Arabia is competing with the UAE, where G42, OpenAI and Oracle are developing Stargate UAE at 1 gigawatt as the first stage of a 5-gigawatt Emirati-American AI campus, with Nvidia as a technology partner. Both countries depend on US export approvals to obtain advanced chips.
• The calendar: Saudi Arabia postponed the Future Investment Initiative conference, scheduled for 26 to 29 October, to next year; Amin had been listed among its speakers. The FII Institute says the conference has helped generate $250 billion in investment agreements and pledges since its launch.

Between the lines

The two-phase structure separates what is contracted from what is conditional: the Blackwell tranche has a deployment quarter, while the far larger Vera Rubin volume still depends on definitive agreements. The unnamed US customer holding 50MW — more than half the announced capacity — also makes the headline figure contingent on a party neither company has identified.

What's next

Watch for the Blackwell deployment to begin in the fourth quarter of 2026, for definitive agreements converting the 70MW Vera Rubin tranche, and for US export approvals governing advanced chip shipments to both Riyadh and Abu Dhabi.

Source: The National