In 1973, when OPEC’s oil embargo revealed how much of the industrialised world ran on Gulf petroleum, the region’s monarchies discovered they had bought themselves a kind of leverage that money alone could not manufacture. Half a century later, the direction of that leverage has reversed. Saudi Arabia and the United Arab Emirates are now the ones writing tens-of-billions cheques for a resource they cannot produce, cannot fully substitute, and cannot obtain without American permission: Nvidia’s AI chips.
The conventional read on the Gulf’s AI spending spree is that vast sovereign wealth, applied ruthlessly, buys technological independence. Buy enough chips, sign enough suppliers, pour enough concrete for enough data centres, and eventually you stop being a customer and start being a peer.
That framing has quietly collapsed over the last twelve months.
What Riyadh and Abu Dhabi are actually building looks less like sovereignty and more like the deepest integration into the US technology stack that any foreign economy has ever attempted. The reason is structural, and it is not going away.
The size of the bet
Saudi Arabia’s Public Investment Fund set up Humain as the vehicle for turning the kingdom into an AI power. Since then, Humain has moved with a speed that would be difficult in any regulated Western jurisdiction. It has secured 211 plots of land across Saudi Arabia for data centres, signed agreements with AMD, Groq, and Qualcomm for computing capacity totaling hundreds of megawatts.
Silicon Canals has previously covered how the kingdom committed $100 billion to AI infrastructure as part of its post-oil diversification push. The UAE has moved on the same scale through G42, whose Stargate facility in Abu Dhabi will run on hundreds of thousands of Nvidia chips.
Read the deal list carefully and a pattern emerges. Humain has bought from AMD, Groq and Qualcomm. It has also bought from Nvidia. The Nvidia order is smaller in raw dollar terms than some of the other deals. It is also the only order that matters for what Humain says it wants to do.
Why the alternatives aren’t really alternatives
The Gulf’s supplier diversification has produced a fleet of chips. The chips do not do the same job.
The distinction is stark: the Qualcomm and Groq silicon is designed to run finished AI models cheaply, for tasks like answering user queries. Training a frontier model — the hard part, the expensive part, the part that decides whether a country has an AI industry or a data-processing industry — still requires Nvidia.
This split between training and inference is now the defining fault line in the chip market. Datacenter Knowledge has described inference as the next AI chip battleground, precisely because it is where Nvidia’s competitors have a shot. Training is where they do not.
The AMD, Groq and Qualcomm deals give Humain the ability to serve models. They do not give Humain the ability to build them from scratch at frontier scale. If a Saudi lab wants to train something competitive with GPT-class systems, the hardware order goes to Santa Clara.
CUDA is the moat, not silicon
The chip itself is only half of Nvidia’s grip. The other half is software.
CUDA, Nvidia’s parallel computing platform, has been under continuous development for nearly two decades. Millions of developers now build within it. Every graduate student who trained a model, every startup that shipped a product, every library and framework and optimisation trick — the accumulated labour of decades sits inside an ecosystem that only runs on Nvidia hardware.
Moving to a rival is not a purchasing decision. It is a rewrite. AlphaStreet has argued that CUDA lock-in combined with supply scarcity makes the Nvidia moat harder to break than the raw silicon comparison would suggest.
Qualcomm’s own strategy tells the story. Silicon Canals has covered Qualcomm’s multi-billion dollar move on a chipless startup, a bet that catching up to Nvidia requires buying software capability rather than fabricating more transistors. The pattern suggests that the silicon race is a distraction from the ecosystem race.
The Gulf’s spending confronts the same reality from the buyer’s side. A megawatt of AMD compute costs less than a megawatt of Nvidia compute. It also produces less usable AI, because the tooling maturity is not there.
China isn’t the escape hatch
Every conversation about breaking Nvidia’s grip eventually arrives at China. Huawei’s Ascend chips, SMIC’s manufacturing push, the theory that a parallel Chinese ecosystem could give the Gulf a second option.
The theory does not survive contact with two facts.
First, Chinese chips trail Nvidia’s by at least a generation. Chinese domestic demand also absorbs most of what China’s foundries can produce, leaving little for export even if export were geopolitically feasible.
Second, and more decisively, the United States has made access to American AI technology conditional on keeping Chinese hardware out. This is not a soft preference. It is written into the export-licensing regime that governs whether Blackwell chips can be shipped to Riyadh at all. The Gulf gets a choice: American chips or Chinese chips, not both.
Given that Chinese chips are worse and American chips are essential for training, the choice makes itself.
Money doesn’t cut the queue
There is a version of this story in which sovereign wealth changes the calculus. The Gulf has more liquid capital than almost any other bidder. Surely, at some point, cash gets you to the front of the line.
The reality appears different. American hyperscalers — Microsoft, Meta, Google, Amazon, plus the AI labs themselves — can access effectively infinite capital at extremely high valuations. Money, at the current margin, is not the scarce input. Foundry capacity is. High-bandwidth memory is. Skilled engineering time is.
The scale of demand shows up in Nvidia’s numbers. The company reported $51.2 billion in data centre revenue in the quarter ending October 2025, up 66% year-over-year, with Blackwell described as sold out. When a multi-trillion dollar company says its flagship product is sold out, sovereign chequebooks are not the tiebreaker.
Even OpenAI, sitting inside the American capital markets and with direct partnerships to Nvidia, has needed tens of billions in fresh funding to keep pace with compute demand. The Gulf’s billions are competing for supply against buyers with equal or greater urgency and equal or greater access.
The upstream bottleneck nobody escapes
Suppose, for a moment, that Humain succeeded in shifting most of its orders to AMD, or Qualcomm, or an ASIC vendor building custom silicon for a specific workload. It would still hit the same wall further down the supply chain.
Nvidia’s chips are fabricated by TSMC. So are AMD’s. So are most of Qualcomm’s, and most of the ASICs. The Taiwanese foundry is the choke point, and its capacity for advanced nodes is allocated years in advance. Buying from a Nvidia rival does not sidestep TSMC. It just changes which customer number you are in the same queue.
High-bandwidth memory is the second bottleneck. HBM3E and its successors come from a handful of Korean and American producers. Everyone building an AI chip needs it. Nvidia gets priority because it is the biggest buyer.
The Gulf’s supplier diversification, in other words, is a diversification of front-end vendors sitting on top of an undiversified back-end.
The political dependency doesn’t diversify
Even the commercial diversification, such as it is, does nothing to change the political geometry. AMD chips, Qualcomm chips and Groq chips are all US products subject to the same export controls as Nvidia’s. Adding them to the shopping list reduces exposure to one vendor’s pricing decisions. It does not reduce exposure to Washington.
The ceiling is clear: true technological sovereignty in AI is, for nearly every country outside the US and China, effectively unattainable. Even Europe, with its industrial base and capital markets, lacks a realistic path to a fully independent AI stack on any near-term horizon.
What the Gulf is doing is the opposite of diversification. It is acceleration of integration. Massive data centres, direct partnerships with US firms, equity stakes in American technology companies. The bet is that being indispensable to the American AI economy is more valuable than being independent of it.
Realism as strategy
There is a case that this is the smart play. The Gulf’s oil rent is finite. The window to convert hydrocarbon wealth into a durable post-oil economy is measured in decades, not centuries. Waiting for a mature non-Nvidia training stack that may never arrive would waste the window.
Silicon Canals has explored how Dubai’s economy has already largely moved past oil dependence, with hydrocarbons now under 1% of the emirate’s GDP. The AI push is Saudi Arabia’s attempt at the same manoeuvre, one economic cycle later and at a much larger scale.
The realism has costs. Every dollar spent building a Blackwell-powered data centre is a dollar spent locking the Gulf deeper into an American supply chain that Washington can throttle at will. The 2019 export controls that kneecapped Huawei showed that the US treats chip access as a foreign policy instrument. The Gulf is choosing to sit inside that instrument’s reach.
The alternative — building an independent stack — is, on current evidence, not available at any price. So the choice is not between dependence and independence. It is between dependence with participation and dependence without.
Where the leverage actually sits
The Gulf is not powerless in this arrangement. Its sovereign funds hold meaningful stakes in the US technology companies it buys from. Its data-centre construction speed — driven by cheap energy, cheap land, and a regulatory environment that says yes — is faster than what US hyperscalers can achieve on American soil.
Nvidia itself has been leaning into this. Fortune reported that CEO Jensen Huang name-checked Humain three times on the company’s November 2025 earnings call — the kind of repetition executives reserve for deals they want investors to model into forward revenue. Those mentions landed a day after Huang joined a White House state dinner for the Saudi Crown Prince, whose first US visit since 2018 doubled as an announcement stage for a Humain–Nvidia–Amazon deployment.
The Gulf, in other words, is now a line item on Nvidia’s earnings script — a customer prominent enough to move the story Wall Street tells about the company’s next five years. That is not sovereignty. It is not independence. But it is the closest thing to structural relevance that a foreign buyer can extract from the American AI stack today: a permanent seat at the table of the firm that decides who gets to build with the fastest chips on the planet.
For a region that started this decade asking what came after oil, that may be as close to a good answer as the current stack allows. Not sovereignty over the machine — participation in it, at a scale nobody outside Washington and Beijing can match.
