According to Reuters, citing a letter it has seen and one source, the US is preparing to tell 35 partner countries that they must pick a side in the AI race. The message is explicit: a country joining China's rival framework will be left outside the US-led Pax Silica initiative.

What is on offer

Pax Silica is the partnership framework the US has built in artificial intelligence. What initiatives of this kind offer usually falls under the same headings: access to advanced chips, joint research programmes, standards alignment and security cooperation.

The new condition ties that package to an exclusion clause. Membership no longer means only "join us" but also "do not join them".

Why it is this blunt

AI infrastructure consists of layers that have to work together: chips, cloud services, models, standards. When a country joins two different frameworks at once, transfers open up between those layers and export controls lose their force.

US restrictions on chip exports are the concrete example. A restricted chip can reach its destination by circulating through a third country. The exclusion clause aims to close that route.

The dilemma for countries

For most of the 35 this is not an easy decision. For a typical mid-sized economy the picture looks like this:

  • The US side — the most advanced chips, the most capable models and the major cloud providers are here.
  • The China side — cheaper hardware, open-weight models and often more flexible commercial terms.
  • Both at once — possible until now; this step closes it.

Closing the third option is the real change. Many countries have until now been able to hold a balance by buying from both.

For mid-sized economies

An ultimatum of this kind turns technology procurement into a component of foreign policy. Which chip a country buys is no longer only a matter of price and performance; it is a declaration of which bloc it stands in.

In the short term that may create bargaining power for the countries choosing — neither side wants to lose one. In the long term, technological dependence locks in a single direction, and that lock does not open easily.

What is not settled

The report describes a plan; no formal notification has yet gone to the countries. The final form of such initiatives usually emerges softer than the first draft, with exceptions added.

The direction is clear regardless: AI is ceasing to be a heading in technology policy and becoming a direct instrument of foreign policy.

It has happened before

The pattern is not new in the history of technology. A similar demand to pick sides played out during the 5G arguments over telecoms infrastructure, and the result was two separate equipment ecosystems.

Two lessons came out of that experience. First, the cost of separation runs higher than forecast: complexity and expense rise for countries and companies forced to support both ecosystems. Second, the separation is never complete — systems kept apart on paper find ways to connect in practice.

Who actually loses

In a division of this kind the parties hurt most are the countries in neither bloc. AI research has been conducted largely in the open until now: papers published, models shared, datasets used in common.

Bloc formation narrows that openness. The losers are the research communities without the capital to train their own models, who have advanced so far by drawing on open sources.