Saudi Arabia and the UAE are spending heavily on AI infrastructure, but Rest of World wrote that their largest disclosed projects still depend on Nvidia chips and U.S.-controlled supply chains.

The account described Gulf buyers as seeking more suppliers while Nvidia remains the main hardware route for training advanced AI models.

The Gulf AI infrastructure Nvidia supply chain problem now sits across data centres, chip approvals and memory capacity.

Rest of World cited Saudi Arabia's Humain, the UAE's G42 and several analysts for a market where capital does not by itself secure immediate access to the most capable AI hardware.

Humain And G42 Keep Nvidia At The Centre Of Gulf AI Buildout

Humain, the AI venture set up by Saudi Arabia's Public Investment Fund, struck a June 1 agreement to use Nvidia technology in self-driving taxis, according to Rest of World.

The Saudi venture is also building data centres in Riyadh and Dammam using several hundred thousand Nvidia chips.

G42 is building the Stargate data centre in Abu Dhabi.

G42 chief executive Peng Xiao described Stargate's machines in a January interview as mostly Nvidia, Rest of World wrote.

Nvidia, G42 and Humain did not respond to Rest of World's requests for comment.

The account records large Gulf AI projects with named buyers and sites, but not final delivery schedules for every chip order.

AMD, Groq And Qualcomm Deals Cover Narrower AI Workloads

Humain has tried to widen its supplier list beyond Nvidia.

Rest of World listed a $10 billion AMD deal for 500 megawatts of computing power, a $2 billion agreement with Groq for chatbot infrastructure and a Qualcomm tie-up for 200 megawatts of chips in a Saudi data centre.

Kamil Dimmich, a partner at North of South Capital, told Rest of World that Qualcomm and Groq chips are aimed at running finished AI models cheaply, while training powerful models still requires Nvidia hardware.

Dimmich told the outlet that Humain's first Nvidia order covered 18,000 Blackwell chips.

Rest of World also cited Nvidia's earnings release for the quarter ending October, when the data-centre business produced a record $51.2 billion in revenue, up 66% from the previous year.

Nvidia described its Blackwell computing line as sold out in that release.

U.S. Approval, CUDA And TSMC Capacity Limit Diversification

Sam Winter-Levy of the Carnegie Endowment for International Peace told Rest of World that buying AMD or Qualcomm chips does not remove political dependence because those products also require Washington's approval before sale to Gulf customers.

Winter-Levy stated that U.S. access to top AI technology is conditional on excluding Chinese hardware.

The technical lock-in is not only the chip.

Rest of World wrote that more than 4 million developers work in Nvidia's CUDA software system, which the company has built for almost two decades to run AI workloads on its processors.

Dimmich told Rest of World that rival chip designs still depend on high-speed memory and manufacturing capacity at TSMC.

Application-specific chips can be cheaper for some inference workloads but still need foundry capacity at the Taiwanese manufacturer, he added.

Gulf AI Sovereignty Remains A Supply-Chain Question

Mohammed Soliman, a senior fellow at the Middle East Institute, told Rest of World that full AI technological sovereignty is effectively unattainable for nearly every country outside the U.S. and China.

Winter-Levy stated that Gulf money, stakes in U.S. technology companies and fast data-centre construction give the region bargaining power, but the current buildout leans further into U.S. partnerships.

Rest of World did not disclose final delivery dates for the Nvidia chip orders, U.S. approval timetables, named alternative suppliers for Stargate's full buildout, or confirmed TSMC capacity allocations for the Gulf projects.