When the chief executive of Saudi Arabia’s state-backed Artificial Intelligence start-up HUMAIN vowed earlier this year to compress two decades of national progress into a single year of AI development, he captured the mood of a region on the move. Within weeks, Washington approved the export of tens of thousands of Nvidia’s most advanced processors to HUMAIN and its Emirati rival, G42, while Riyadh raised its pledged investment in the United States to $1 trillion. Abu Dhabi’s gigantic Stargate UAE infrastructure cluster is rising, as well, and Doha has entered the race through its own sovereign vehicle, Qai. The message to the Gulf public is confident and consistent: This is the foundation of a sovereign, post-oil future, the moment the region stops importing its tomorrow and begins producing its own.
However, it is worth being precise about what is being purchased, because it is not sovereignty.
Sovereignty is not simply the possession of capabilities. It is also the capacity to make consequential choices without another party’s permission. A state can hold enormous power yet lack sovereignty, if the terms on which it wields that power are set elsewhere. Measured that way, most of the AI infrastructure the Gulf has built to date is not an engine of independence but an architecture of dependence: sophisticated, expensive, and pointed in the wrong direction.
Consider the projects underway. The chips are American. The cloud that runs them is, in most of these deals, operated by American firms. The frontier models—large, cutting-edge AI systems—are American. One careful analysis describes the Gulf AI arrangements not as a single model but as models on a spectrum: At one end, enclaves inside G42’s own facilities are run by Microsoft under U.S. license conditions; at the other, there is Nvidia-HUMAIN deployment in Saudi Arabia, which is genuinely sovereign but still relies on U.S. chips. The uncomfortable truth is that the center of gravity sits near the dependent end—the Gulf states as host nations, foreign hands on the controls.
Then consider the price of admission. To win Washington’s approval for the imports described above, G42 tore out its Chinese technology, adopted screening protocols that exclude Chinese firms, and divested from its Chinese holdings. Read that again. The U.S. did not merely sell chips to the Gulf; it bought the right to decide whom the Gulf may partner with. Sovereignty does not come bundled with a supplier’s list of the counterparties the buyer may no longer touch. That clause is one tell; others lurk in the monitoring and reporting requirements written into the licenses, and the operational control the American operators retain. Whoever runs the enclave—sets the conditions and operates under an export authority overseas—holds a lever the host does not. The question is not how much computing power the Gulf has bought; it is who holds the off-switch.
The Limits of Dependency as Security
Now the strongest case for the defense, because it deserves one. Perhaps the Gulf knows exactly what it is doing. By conceivably spending more than a trillion dollars in the American economy and turning its own territory into load-bearing capacity for U.S. tech giants Microsoft, Google, Amazon, and Nvidia, the region may be practicing a deliberate strategy of converting commercial entanglement into a security guarantee. In other words: make Gulf stability indispensable to American balance sheets, and the Gulf states’ adversaries keep Washington as an adversary, too. This spring’s war between the U.S., Israel, and Iran, which included Iranian drone strikes on the Gulf’s own digital and energy infrastructure, was the first live test of that proposition, and the alignment held.
But an insurance policy you do not control is not a policy. It is a promise. The same entanglement meant to protect you is also a leash, and dependency-as-security lasts precisely as long as the relevant parties’ interests align. The Gulf has seen this film before. The oil-for-security bargain struck aboard the USS Quincy in 1945 and sealed in the petrodollar arrangements of the 1970s made the region rich and courted, but the terms were never the region’s to set. When Riyadh’s priorities diverged from Washington’s over the October 2022 oil production cuts, the response arrived within days: talk of “consequences,” a revived NOPEC bill, and arms transfers under review. And the technology remake has screened once already. In 2021, Washington made the F-35 conditional on Abu Dhabi tearing Chinese firm Huawei out of its networks; the UAE refused the price then, and has paid its equivalent now, for chips. A relationship that is leverage today becomes a limit tomorrow, on what you may export, on whom you may deal with, on what you are allowed to build. Handing the stronger party the ability to write those terms is not an end to dependency. It is dependency in its most advanced form.
The contrast with Europe is instructive. The EU, having discovered that roughly four-fifths of its digital infrastructure depends on non-European providers, now treats digital sovereignty as an explicit strategic project, restricting American cloud services for sensitive public data, funding home-grown alternatives, and legislating to seize the control it does not yet possess. One can debate whether Brussels will succeed, but what is not in doubt is the direction. Europe is straining to climb out of exactly the dependency the Gulf is paying a fortune to climb into.
None of this makes the bet foolish or the buildout wasted. It simply means the Gulf should name its strategy accurately and act on the difference. Real sovereignty is not rented. When it comes to AI, that sovereignty lives in the layers a state can actually own: its data, which is the raw material of every future advantage; and portable, open-weight models it can run and adapt without asking permission. Above all, sovereignty lives in the state’s people. The region’s true constraint is not capital—it has more than anyone—but the lack of engineers and researchers who can build and maintain frontier systems without foreign hands, a pool still far too shallow for the ambition. The genuinely sovereign end of the spectrum, the UAE’s open model releases and Saudi Arabia’s Arabic-language systems, is where the real future lies. Today it is the exception. It should be the whole strategy.
The buildout is real, and it is not stopping. But capacity bought on someone else’s terms is not independence; it is a more sophisticated version of the dependence the Gulf was meant to end. The test is not what gets switched on this year; it is who can switch it off in 10 years. Until Riyadh and Abu Dhabi control the switch themselves, they are not building sovereignty: they are financing someone else’s.
