The Gulf has billions to spend on AI. It still needs Nvidia

Money, it turns out, doesn't buy you out of Nvidia. Saudi Arabia and the UAE have spent the last year trying to spread their chip orders across as many suppliers as possible, hoping to avoid depending on one company for their AI ambitions. Saudi's Humain, backed by the kingdom's Public Investment Fund, struck a self-driving taxi deal on June 1 that runs on Nvidia hardware and software end to end. The UAE's G42 is building an Abu Dhabi data center called Stargate that will run on 400,000 Nvidia chips in its first phase alone. CEO Peng Xiao admitted back in January that despite months of trying to diversify, the machines inside Stargate would still be mostly Nvidia.

The numbers explain why walking away isn't really an option. Nvidia's data center business pulled in $51.2 billion in the quarter ending October, up 66% from a year earlier, and its newest Blackwell chips are, in the company's own words, sold out. Part of that dominance is the silicon. Part of it is CUDA, the software platform Nvidia has spent almost two decades building, which now has more than 4 million developers working inside it. Switching to a competitor doesn't just mean buying different hardware — it means rebuilding years of engineering work from the ground up.

Humain has still tried. It signed a $10 billion deal with AMD for 500 megawatts of computing power, a $2 billion deal with Groq for chatbot infrastructure, and a partnership with Qualcomm for 200 megawatts of chips in a Saudi data center. But these deals fill a narrower gap than people assume. Kamil Dimmich of North of South Capital points out that Qualcomm and Groq chips are built to run finished AI models cheaply for everyday tasks, while training a powerful model still requires Nvidia's heaviest hardware — something Humain confirmed when its first Nvidia order was 18,000 of the newest Blackwell chips, tech no rival currently matches.

And escaping Nvidia through a different American company doesn't really escape Washington either. AMD and Qualcomm chips need the same U.S. export approval Nvidia's do, according to Sam Winter-Levy of the Carnegie Endowment for International Peace. China could theoretically offer an alternative, but its best chips lag Nvidia by at least a generation, its own firms buy up most of the supply, and the U.S. has made Gulf access to its top chips conditional on excluding Chinese hardware entirely. Even picking a non-Nvidia chip doesn't dodge the bottleneck further down the supply chain, since rivals compete for the same scarce high-speed memory and the same limited manufacturing slots at TSMC.

What's striking is that even bottomless Gulf wealth doesn't guarantee a shortcut. Dimmich notes that the U.S. tech giants competing for the same chips can raise just as much capital, so cash alone isn't what puts you at the front of the line. Mohammed Soliman of the Middle East Institute goes further, arguing that true technological independence in AI is so expensive that it's essentially out of reach for everyone except the U.S. and China — even Europe, he says, can't realistically build a fully independent AI stack anytime soon. Winter-Levy still credits the Gulf's money, its stakes in U.S. tech firms, and its speed in building data centers with giving it genuine leverage. But the states appear to be spending that leverage on getting closer to Nvidia and the U.S., not on breaking free of them.