Last week I argued that AGI, like fire, will not reach everyone equally. This week, the US government proved the point.

On Friday evening, Amazon Web Services revoked access to Anthropic’s two most advanced AI models, Fable 5 and Mythos 5, across every region, for every user outside the United States. No advance notice. No negotiation. A US Commerce Department directive had arrived, Anthropic had complied, and AWS had executed. The whole thing was over before most of the affected world had logged on for the weekend.

That last detail matters more than it has been reported to matter.

Amazon is not a neutral plumbing provider in this story. It is a major investor in Anthropic. It also publicly backed US legislation to restrict AI chip exports, a position widely read as a strategic move to concentrate compute supply in fewer, friendlier hands. And it runs the cloud layer through which Fable 5 and Mythos 5 reached the world. The company that profits from Anthropic’s global reach was also the company that switched off that reach on government instruction, in every region, simultaneously.

This is what a controlled chokepoint looks like when it is activated for the first time.

Three layers, one dependency

The standard framing of AI sovereignty focuses on chips. Who makes them, who is allowed to buy them, what happens when that access is cut. That framing is already dated.

The Fable 5 episode reveals that the dependency runs three layers deep. Compute: chips and the data centres that house them, concentrated in US hands. Intelligence: frontier models trained on that compute, concentrated in a handful of American companies. And delivery: the cloud infrastructure that makes those models accessible to the world, concentrated in three American firms, of which Amazon is the largest.

India has no meaningful presence at any of these three layers at the frontier level. An Indian startup building a vernacular health diagnostics tool on Anthropic’s API did not lose access because of anything it did. It lost access because a directive issued in Washington moved through Amazon’s servers and reached it before New Delhi had been consulted, informed, or given the opportunity to object.

The muted response is the story

What is striking is how little alarm the suspension generated outside technology circles. When the US placed Huawei on its entity list, it was front-page news across every major economy for weeks. When AI model access is suspended for all foreign nationals on national security grounds, the response is a few columns and some LinkedIn commentary.

The muted response reflects a failure of instinct. Governments and publics have not yet learned to recognise, in real time, that something geopolitically significant has occurred when a cloud provider revokes model access across regions on government instruction. That instinct needs to develop quickly, because the policy infrastructure to do this again, to more models, with more speed, now demonstrably exists.

What Amazon’s position reveals about India

Amazon’s dual role as Anthropic investor and access enforcer is not a contradiction. It is a clarification. The commercial interests of American cloud providers and the national security interests of the American government are, at this layer of the stack, largely aligned. Both benefit from frontier intelligence remaining dependent on US-controlled infrastructure. India sits entirely outside that alignment.

Reports suggest India is among Anthropic’s largest international markets. That means Indian developers, researchers, and enterprises represent a significant share of the user base that was cut off. The suspension was not aimed at India. But India had no seat at the table when it happened, no prior agreement that would have exempted Indian institutional users, and no domestic alternative to redirect to.

This is digital sovereignty measured not in policy documents but in what actually occurred on a Friday evening when a government directive moved through a cloud provider’s infrastructure.

India cannot afford to be a bystander

MeitY needs to be negotiating AI access agreements with Washington now, not after the next suspension. Not as a supplicant asking for exemptions, but as a major market demanding institutional standing in decisions that directly affect Indian developers, researchers, and enterprises. The civil nuclear deal and the defence technology frameworks show that such agreements are possible when India treats access as a strategic interest rather than a courtesy extended by a partner.

The deeper work is harder. India’s AI sovereignty investments need to be evaluated against all three layers this episode exposed, not just models, where attention currently sits, but cloud delivery, where the enforcement actually happened. As long as the off switch is in someone else’s data centre, the question of who controls Indian AI infrastructure has a simple and uncomfortable answer.

Intelligence with borders is not an abstraction. It is a sovereignty problem, and India is not at the table where it is being solved.

END OF ARTICLE