The Gulf has spent the past year buying its way to the front of the AI queue. On Friday it was reminded who controls the queue. Anthropic said it had received a US government directive and, within hours, cut off its two most capable models, Fable 5 and Mythos 5, for every foreign national. Al Jazeera reported that the company was handed only partial information and no detailed security rationale. For Abu Dhabi and Riyadh, where sovereign funds have wagered billions on becoming AI hubs, the order lands as a precise illustration of the dependency their strategies are meant to reduce.
The detail matters. The phrase every foreign national does not mean users outside the United States. It means anyone who is not a US citizen, wherever they are, which on its face captures every engineer, analyst and civil servant in the region who had built on those models. Fortune and CNN reported that the concern, as Anthropic understands it, involves a possible way to bypass a safeguard so that Fable 5 could help identify software vulnerabilities. Anthropic complied while disputing the order, warning that a standard this broad would stop the release of any new frontier model and noting that rival American systems carry similar capability.
The Gulf's bet, stated plainly
The numbers behind the region's ambition are large and recent. In November 2025 the US approved sales of advanced Nvidia chips to the UAE's G42 and Saudi Arabia's Humain, Bloomberg reported, authorising each to acquire computing power equivalent to roughly 35,000 of Nvidia's most advanced processors, worth about a billion dollars apiece. CNBC tied the approval to the Saudi Crown Prince's Washington visit and to Gulf commitments to limit Chinese technology in their stacks. Humain is wholly owned by the Public Investment Fund, which manages close to a trillion dollars.
That capital has been moving fast. G42's chief executive said the first chip shipments would arrive within months to support an initial 200 megawatts of capacity for a planned Stargate cluster. Gulf News framed the approvals as the moment Gulf AI ambition moved from commercial plan to geopolitical fact. The region has the compute, the power and the money. Friday's directive is a reminder that it does not yet control the layer that sits on top of the silicon: the frontier models themselves.
Export controls cut both ways
The Gulf understood the chip deal as a grant of access in exchange for alignment, keeping Chinese kit out in return for American hardware. The Anthropic order shows the same logic applied to software, and it is less forgiving. Chips, once installed, keep working. Model access delivered over an API can be withdrawn on the day a directive is signed, with no notice and no appeal. A sovereign AI programme that runs the region's best applications on a US model has built its flagship on rented ground.
This sharpens the case the region's own strategists have been making. Saudi Arabia and the UAE have funded Arabic-language model work and national labs precisely so that critical workloads do not depend on a single foreign supplier. Until Friday that argument competed with the convenience of simply calling an American API that already worked. The convenience just acquired a visible risk premium. When the supplier can gate by passport, the locally controlled model stops being a matter of national pride and becomes a continuity requirement.
There is a harder strategic reading too. The Gulf's compute build-out is being financed partly to serve others, including American firms running clusters in the region. A directive that bars foreign nationals from the most capable models raises an awkward question about who may operate, and who may merely host. The region wants to be a principal in the AI economy, not a landlord for it. Controlling the model layer is the difference between the two.
What Abu Dhabi and Riyadh do on Monday
The immediate task is an inventory. Which government and enterprise systems depend on Anthropic's models, and which have a tested fallback to another provider or to a locally hosted open-weight system. The medium task is contractual: continuity terms, multi-model commitments and clarity from American suppliers about how a future directive would be handled. The long task is the one the sovereign funds already started, building model capability the region can run on its own compute and switch on without asking permission.
The Gulf is better placed than most to absorb this. It has the chips arriving, the power to run them and the balance sheets to fund the labs. What Friday changed is the urgency. A model that can be revoked by nationality is not infrastructure a sovereign AI strategy can lean on, however good it is. The region bought its way to the front of the queue. Now it has a reason to build a queue of its own.
The Arabic-model case just got stronger
The region already has the beginnings of an answer, and Friday makes it look less like national pride and more like prudent engineering. The UAE's work on the Falcon family, Abu Dhabi's Jais models for Arabic, and Saudi Arabia's Allam system were funded on the argument that a Gulf that cannot reason in its own language and on its own infrastructure is not truly sovereign in AI. For two years that case competed with the simple fact that an American API was better and easier. The order narrows that gap by adding a risk the American option carries and the local one does not, namely the chance of being switched off by a government that is not your own.
The same logic reaches the region's role as a host. A meaningful part of the Gulf's compute build-out is intended to serve others, including American firms placing clusters in Abu Dhabi and the Saudi desert to draw on cheap power and fast approvals. A directive that bars foreign nationals from the most capable models complicates that arrangement, because it raises the question of who is permitted to run what on hardware the region paid for. The Gulf wants to be a principal in the AI economy. Being a landlord whose best tenants answer to Washington is a weaker position, and Friday made the distinction visible.
There is also a timing point the region's planners will note. The chips arrive within months, the Stargate capacity comes online through 2026, and the question of what software runs on all that silicon is suddenly less settled than it looked in November. Sovereign compute waiting on a foreign model is an expensive form of dependence. The push toward Arabic frontier models, regional model-hosting agreements with clear continuity terms, and open-weight systems the state can run without permission has moved from the slideware of strategy decks into the operational column.
The chip deals of November were the easy part, capital meeting hardware. Friday is the part that tests the strategy. The Gulf's ambition was always to own AI capability rather than rent it, and an order that gates the best models by citizenship turns that ambition from aspiration into deadline. Keep buying the compute, keep funding the labs, but stop treating any single American model as load-bearing. Sovereignty that can be switched off from Washington is not sovereignty. It is a subscription.
