Pax Silica
Thirty-five countries signed something called the AI Opportunity Statement. This week they found out it was an exclusivity agreement.
Reuters reported on August 18 that the State Department has drafted a letter to those signatories telling them they have to choose: the American AI coalition, reported to be carrying the name Pax Silica, or Beijing’s competing framework. Not both. Countries that join China’s bloc risk being pushed out of the US-led one.
The next day, China’s foreign ministry spokesperson Lin Jian said Beijing opposes “taking sides and forming camps” on AI, and that every country has the right to choose its partners based on its own national conditions and development needs.
Both statements are about sovereignty. Only one of them is credible in the mouth of the government saying it, and it isn’t obvious which.
What a stack becomes when you make it a loyalty test
For about a decade the working assumption in technology policy was that the AI stack was a market. You bought chips from whoever made the best ones, rented compute from whoever had capacity, and downloaded weights from whoever published them. Nationality was a supply-chain risk, not an identity. You could run Qwen on Nvidia silicon in a European data center and nobody thought you had declared anything.
The draft letter ends that. Not because it restricts a specific technology — export controls have done that for years, and export controls are a normal instrument of statecraft. It ends it because it attaches a consequence to association. The question stops being what you deploy and becomes who you deploy it with.
Carl Schmitt’s claim about politics was that the properly political distinction is friend and enemy, and that it doesn’t require hatred, or even conflict. It requires only that the possibility of conflict is real enough that the grouping matters. Everything else — economics, ethics, aesthetics — becomes political at the moment it starts sorting people into those two piles.
A technology stack has now been sorted into those two piles. That is a bigger change than any single control on any single chip.
Schmitt is uncomfortable company, and I’m not reaching for him because he was right about how to run a state. He was catastrophically wrong about that. He’s useful here for the narrow descriptive reason that he saw the mechanism clearly: the political isn’t a subject area, it’s an intensity. When association with a vendor can cost you membership in a coalition, the vendor relationship has reached that intensity. Buying inference is now a foreign-policy act.
The word both sides are using
The interesting thing about the exchange is that “sovereignty” is doing the load-bearing work on both sides, in two incompatible senses.
Beijing’s version is sovereignty as non-interference: each state chooses its own partners, and no third party gets to price that choice. This is the old Westphalian formula, and it is the same formula China has used for two decades on trade, on internet governance, on Xinjiang, on Taiwan. It is rhetorically powerful and it is being deployed by a government that maintains one of the most aggressively controlled national information environments on earth. The principle is real. The messenger has a record.
Washington’s version — mostly implicit, since it lives in a draft letter rather than a speech — is sovereignty as capability: you are only meaningfully independent if the substrate you run on is one you can trust and, at some level, influence. On this reading, a country that builds its public services on a foreign model it cannot audit hasn’t preserved its sovereignty by choosing freely. It has spent it.
Both are coherent. They are also both self-serving, and the countries receiving the letter can see that from a long way off.
What neither version answers is the question a mid-sized democracy actually has to answer this year: what is left of your independence when the choice is between two stacks you don’t control, and the price of neutrality is exclusion from the one with the compute?
The third-stack argument, and why it’s harder than it sounds
There’s a live argument — Brookings has been making a version of it — that the right response to a forced binary is to refuse it and build a third stack: an open, jointly governed alternative that lets middle powers cooperate without joining either bloc.
The historical rhyme is Bandung in 1955, where twenty-nine African and Asian states met and produced the non-aligned movement out of exactly this predicament. Non-alignment was never neutrality in the passive sense. It was a claim that the bipolar frame was itself the imposition, and that refusing the frame was a form of power.
It worked, partly, and it was also always weaker than it sounded, because non-alignment during the Cold War was possible in a domain — ideology — where the barrier to entry was a printing press and a conference hall. The barrier to entry on the current frame is a fab, a grid, and about forty billion dollars.
That’s the asymmetry that makes the third-stack argument hard. You can decline the friend/enemy grouping on the level of rhetoric for as long as you like. Declining it materially requires a supply chain, and there are two of those, and the frontier of both is controlled by a handful of firms inside the two states writing the letters. Europe has the capital and the regulatory apparatus and, at present, neither the silicon nor the power. India has the talent and the demand and is building the rest at speed. Nobody has all four.
The non-aligned position is available. It is just expensive in a way non-alignment was not in 1955, and the countries most attracted to it are the ones least able to pay.
What the letter admits
The thing worth sitting with is not that the US is applying pressure. Great powers apply pressure; that’s the job.
It’s what the pressure concedes. You do not need to make people choose between two things unless the second thing is a real option. The letter is an admission that the Chinese stack has become good enough, cheap enough, and available enough that a sovereign state might rationally pick it — which is precisely the situation Hugging Face’s own download numbers describe, and precisely the situation export controls were built to prevent.
Coalitions get walls when they stop being obviously the better deal.
The signatories will mostly sign. Alignment is cheaper than the alternative, and for most of the thirty-five the compute is on one side of the line and nothing on the other side changes that. But the frame will hold longer than the letter does, and the frame is the durable thing here — because once a technology has been sorted into friend and enemy, every subsequent decision about it inherits that sorting, including the ones nobody meant to politicize.
The chips were always going to be a strategic asset. What’s new this week is that the choice became one.
Sources: Reuters via The Next Web on the draft letter to 35 countries ↗ · China urges respect for digital sovereignty, August 19 2026 ↗ · China blasts US plan of asking countries to take sides, China Daily ↗ · Making the case for a third AI technology stack, Brookings ↗ · The sovereignty gap in U.S. AI statecraft, Lawfare ↗