The AI Walls Are Producing Independence, Not Allegiance

Two walls are going up at once, and they are mirror images. The United States is denying foreign nationals access to American AI models — Anthropic cut off Indian engineers this spring under export controls ^[1]. China, as of today, is drafting national-security laws to wall in its own model weights, training data, and chip designs, making their overseas transfer a criminal offense ^[2]. Both blocs are now doing the same thing: treating AI as a strategic asset to be contained, not shared. The mechanism this creates is straightforward, and India states it most explicitly.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. — Anthropic

Deny a middle power access to frontier models, and you do not keep it dependent — you make indigenous build a necessity rather than a choice. The wall produces the very thing it was meant to prevent: a competitor outside the gate. The breadth of the response is what turns a mechanism into a pattern. In February, Canada and Germany launched a Sovereign Technology Alliance with an explicit purpose.

Germany and Canada are united in our belief that responsible AI development and resilient digital ecosystems are critical to long-term economic strength. — Karsten Wildberger

They named both superpowers as the dependency to escape ^[3]. In May, the European Union unveiled a technological sovereignty package restricting foreign cloud providers from defense, healthcare, banking, and energy, with officials warning that depending on a single dominant supplier — one company or one country — is an unacceptable risk ^[4]. Japan's SoftBank-NEC-Sony-Honda consortium, backed by up to a trillion yen in public money, is building domestic foundation models for technological independence ^[5]. Even Ukraine — a close U.S. ally at war — is building sovereign AI infrastructure, with its telecom CEO arguing that the country's most critical data and capabilities must stay at home ^[6]. And in February, India, Italy, and Kenya signed a trilateral AI pact combining digital public goods, industrial expertise, and regional leadership — an institutional track that belongs to neither Pax Silica nor WAICO ^[7]. These are not examples of alignment with a bloc. They are examples of escape from both. Now the paradox at the center. China's anti-hegemony message is being adopted. At the UN Global Dialogue in Geneva, India and China jointly called for inclusive global AI governance, with China's envoy asserting that digital sovereignty means countries can independently choose AI products without being coerced into taking sides ^[8]. A U.S. congressional commission found that roughly 80% of American AI startups now use Chinese open-source base models, and Alibaba's Qwen has surpassed Meta's Llama in downloads ^[9]. China's Global Governance Initiative, released in June, claims support from over 150 countries and positions the UN — not U.S.-led frameworks — as the primary channel for AI governance ^[10]. But every middle power that takes up the critique uses it to justify independence from China too. The same European Union that invokes sovereignty against American cloud providers is preparing trade restrictions against China, requiring companies to diversify to at least three sources for critical supplies ^[11]. Canada and Germany named both. The message travels; the membership does not. The U.S. is not passively losing this contest. It is competing for middle-power allegiance with concrete access — elevating the UAE to A5 export status, a tier normally reserved for NATO allies, and granting nine American firms unlicensed access to the Emirati market ^[12]. Pax Silica, launched in June, convenes 35 nations to secure the AI value chain against China ^[13]. Both superpowers are organizing. Neither is capturing the middle. And the U.S. is widening the vacuum it refuses to fill. At the New Delhi AI summit in February, White House AI advisor Michael Kratsios was unambiguous.

We totally reject global governance of AI. — Michael Kratsios

That rejection of the multilateral forum is precisely the space China's WAICO and Global Governance Initiative are designed to occupy ^[14]. But the U.S. compounds the problem with a moral binary that leaves no room for middle-power autonomy: Defense Secretary Hegseth said he would not hesitate to reject AI models that cannot fight wars ^[15], and Congressman Brian Mast described the contest as superhero versus supervillain ^[16]. A world divided into heroes and villains has no seats for the undecided. Singapore's minister Josephine Teo put the middle-power position in a single sentence.

The choice is not so much between country A and country B. The choice is really whether, on balance, you are better able to achieve the combination of performance and affordability, security and resilience. — Josephine Teo