1. [para. 1][para. 2][para. 3][para. 4] Over the past year, two rival AI architectures have emerged. In December 2025, the United States launched Pax Silica, a State Department-led coalition meant to secure supply chains for critical minerals, semiconductors, and AI infrastructure. In July 2026, China gathered 29 countries in Shanghai to establish the World Artificial Intelligence Cooperation Organization (WAICO), pitched at the Global South. Neither is yet a settled bloc, but Washington has reportedly warned signatories of its AI Opportunity Statement that they cannot belong to both: “to be part of everything is to be part of nothing.”
2. [para. 5][para. 6][para. 7][para. 8] The more useful question is not whether AI will break in two, but where a break could actually hold, since the answer varies across the technology. The internet and telephone systems fragmented at the surface — through China’s Great Firewall, data-localization rules, and national controls — but never at the base, because networks gain value with reach. AI may not obey the same rule; whether it splits depends on which layers a single actor can control, and some key parts of the stack are already bifurcating between the U.S. and China.
3. [para. 9][para. 10][para. 11] The hardware layer can be split. The U.S., which long dominated this layer, has tried to ring-fence it through export controls. Advanced chips, the lithography that makes them, and high-bandwidth memory have become essential chokepoints in the U.S. quest for AI dominance. Pax Silica coordinates these chokepoints among members that hold such power, while also managing access to chips outside export controls and shared compute infrastructure.
4. [para. 12][para. 13][para. 14][para. 15] The software/model layer cannot be split in the same way. Open-weight models such as DeepSeek, Qwen, and Kimi are already downloaded and run worldwide, and once released, their weights cannot be recalled. Restricting use would require policing every developer and enterprise; sanctioning distillation — training a cheaper model on a stronger one’s outputs — would mean detecting a process that leaves almost no trace. Even cutting off access to proprietary frontier systems, as one leading U.S. developer briefly did this year, does not seal the border because capability leaks through open weights and distillation. Chips can be counted and stopped; model capabilities diffuse, and no export regime can hold them.
5. [para. 16][para. 17][para. 18] The realistic outcome is an asymmetric split: a genuine fracture at the hardware and infrastructure level beneath a software layer that remains porous. This is the worst of both configurations — it fragments the market into two chip ecosystems, two sets of standards, and duplicated infrastructure, without delivering the security either side seeks. The push comes from both directions: Washington is turning a supply-chain coalition into a loyalty test, while Beijing builds WAICO as a permanent body speaking for the Global South, with Xi Jinping’s call for an order in which “no country should dominate” serving as its own bid to lead.
6. [para. 19][para. 20][para. 21][para. 22] Singapore illustrates the cost. A founding member of Pax Silica, Singapore was assured in May by the Trump administration that joining implied no obligation to choose between China and the U.S. By August, it received a letter instructing it to choose. Singapore cannot choose: banning Chinese models would hollow out its own economy and, given the porous software layer, would not even work. It is a trusted partner discovering both that the terms changed after it signed and that the demanded choice is unenforceable.
7. [para. 23][para. 24][para. 25][para. 26] Europe is in the same position, with more at stake. Its sovereignty legislation — the Cloud and AI Development Act and the wider June 2026 technology package — was conceived as a strategic-autonomy bet to avoid choosing. Yet in the same month, the EU signed the Pax Silica Declaration, aligning with the U.S. camp. Autonomy and bloc membership coexist only until someone forces the question. Europe differs from Singapore, though, because it is indispensable for advanced lithography, without which neither side can build advanced chips. That leverage lasts only if Europe uses it as an EU-level tool rather than a Dutch one, since ASML is based in the Netherlands. For now, Europe is relinquishing that leverage to the U.S. virtually for free.
8. [para. 27] The other risk is that China cracks advanced lithography, making the current chokepoint disappear — which would only deepen the bifurcation of AI hardware.
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新