The Gulf’s oil-rich states, betting billions on reinventing themselves as centers of artificial intelligence, are discovering that money can buy almost everything, except a way out of Nvidia.

Saudi Arabia and the United Arab Emirates have committed tens of billions of dollars to becoming AI powers before the oil money fades. For the past year, they have tried to buy chips from as many companies as they can, so that no single supplier controls how much they pay and how fast they can build.

But they are running into a wall, because the few alternatives to Nvidia are either far weaker or, in China’s case, blocked by the U.S.

“Diversification away from Nvidia reduces the Gulf’s dependence on any one commercial vendor, but it doesn’t change the political risk of dependence on the U.S. at all,” Sam Winter-Levy, a fellow at the think tank Carnegie Endowment for International Peace, who studies the geopolitics of chips, told Rest of World. “What would change that is Chinese supply, but China cannot yet export competitive chips at scale, and the U.S. has made the Gulf’s access to U.S. AI tech conditional on keeping Chinese hardware out.”

Saudi Arabia agreed on June 1 to use self-driving taxis that run on Nvidia technology — from the computer inside each car to the software that steers it. The deal with the U.S. chipmaker was struck by Humain, the venture set up by Saudi Arabia’s Public Investment Fund last year to make the kingdom a global AI power. Humain is also building data centers in Riyadh and Dammam, powered by several hundred thousand Nvidia chips.

The UAE is investing on the same scale through its state-backed AI firm G42, which is building a data center in Abu Dhabi called Stargate. Its first stage will run on 400,000 Nvidia chips. G42 has spent months trying to spread its orders among several suppliers, yet the machines inside Stargate would be “mostly” Nvidia, CEO Peng Xiao said in an interview in January.

Nvidia, G42, and Humain did not respond to requests for comment.

The scale of Nvidia’s hold was evident in the quarter ending October, when its data center business brought in a record $51.2 billion, up 66% from the previous year. Its newest Blackwell computing architecture line is “sold out,” Nvidia said in the earnings report.

Part of that hold is the chip itself, and part is the Compute Unified Device Architecture, Nvidia’s proprietary parallel computing platform. Nvidia spent almost two decades building the tools engineers use to run AI on its chips, and more than 4 million developers now work in the CUDA system. Moving to a rival would mean rebuilding years of work from scratch.

Rivals fall short 

Humain has tried hard to widen its list of suppliers. Last August, it struck a $10 billion deal with AMD for 500 megawatts of computing power, a $2 billion agreement with Groq to run its chatbot, and a tie-up with Qualcomm to fill a Saudi data center with 200 megawatts of its chips. 

The deals give the Gulf chips that can run AI, but for a narrower set of tasks than those from Nvidia. The Qualcomm and Groq chips are built to run finished AI models cheaply, for everyday tasks like answering a user’s questions, while training a powerful model still calls for Nvidia’s chips, Kamil Dimmich, a partner at emerging-markets investment firm North of South Capital, told Rest of World.

Humain’s first order from Nvidia was 18,000 of its newest Blackwell chips — the hardware that does that heavy lifting, which no rival can yet match, he said.

Chips from AMD or Qualcomm come with the same strings attached as those from Nvidia, because all of them are U.S. products that need Washington’s approval before they can be sold to the Gulf, Winter-Levy said.

China’s best chips trail Nvidia’s by at least a generation, and its own companies buy up most of what it makes. In addition, the U.S. has told the Gulf that access to its top chips depends on shutting Chinese equipment out, Winter-Levy said. 

Buying from a rival does not sidestep Nvidia in any case, because its competitors depend on the same suppliers further down the chain, Dimmich said. The two hardest things to secure are a specialized high-speed memory that keeps the chips fed with data, and space in the factories run by Taiwanese contract manufacturer TSMC, which makes chips for Nvidia and its rivals alike.

“Nvidia chips are essential for training,” Dimmich said. “For this type of inference compute, increasingly there are ASIC [application-specific integrated circuit] designs that are cheaper than Nvidia but still need foundry capacity at TSMC.”

ASIC designs are chips built for a single job, cheaper than Nvidia’s all-purpose processors.

Billions aren’t enough

Even the Gulf’s great wealth buys less of an edge than it might expect, Dimmich told Rest of World, because the deep-pocketed U.S. technology giants it competes with for chips can raise just as much money. Cash is not what decides who gets to the front of the line, he said.

“I don’t think infinite money is much of an advantage at the moment,” said Dimmich. “All the U.S. buyers of this equipment seem to be able to access infinite amounts in capital markets at extremely high valuations, so that’s not been a constraint, although it may become so in future.”

Building an AI industry beholden to no one has grown so costly that even the world’s richest economies cannot manage it, according to Mohammed Soliman, a senior fellow at the Middle East Institute, a foreign-policy think tank in Washington.

“True technological sovereignty in artificial intelligence is extraordinarily expensive and, for nearly every country outside the United States and China, effectively unattainable,” Soliman told Rest of World. “Even Europe, despite its industrial capacity and capital markets, lacks the realistic prospect of building a fully independent AI stack within any meaningful time frame.”

Still, the Gulf’s money, stakes in U.S. technology companies, and speed in building data centers give it real bargaining power, Winter-Levy said. However, it is leaning into Nvidia to buy the best chips that the whole industry is built on, rather than holding out for alternatives that may never match it. Its leaders, Soliman said, know exactly what they are choosing.

“The Gulf’s current path is the opposite of diversification,” he said. “They are accelerating integration through massive data centers and direct U.S. partnerships rather than chasing expensive and uncertain independence. The Gulf states are AI realists.”