The Chip Coalition Didn't Break. It Was Replaced.

The Netherlands had an American accusation in hand and chose not to use it. The US flagged ASML for illegally shipping EUV technology to China. Dutch Foreign Trade Minister Sjoerdsma declined to initiate official action, calling the existing rules already stringent enough:

The system is incredibly robust and the export rules as they currently stand are extremely strict, perhaps the strictest in the entire world. — Sjoerd Sjoerdsma

ASML denies the charge; the Netherlands has not moved ^[1]. That refusal is where the enforcement coalition stopped functioning — not with a public rupture but with a partner declining to act on the accusing country's own evidence. Six weeks later, 35 nations signed the "Joint Statement on AI Opportunity" — Pax Silica ^[2]. Its architect, Nathan Helberg, framed the logic explicitly against the enforcement model: who builds first, not who regulates first ^[2]. The framework's actual tools are a $50M cargo-verification platform called Pax Pass and a workforce partnership with Stanford called Foundry School — consultative instruments, not binding export controls ^[2][3]. What replaced the coalition cannot enforce what the coalition did. Unilateral enforcement persists, but on a residual track. The Pentagon's 1260H blacklist expanded from roughly 130 to 188 Chinese firms on June 8, adding Alibaba, Baidu, Tencent, and BYD ^[4]. Days earlier, Commerce's BIS closed the overseas-subsidiary loophole, clarifying that licensing requirements apply to all China-headquartered companies regardless of subsidiary location ^[5]. China's Ministry of Commerce condemned the move as tightening ^[6]. Taiwan is considering strict AI chip export bans aligned with US measures ^[7]. But these are unilateral or peripheral allied actions. The core bilateral enforcement relationship — the US-Netherlands axis on ASML — has fractured. The coalition architecture that gave controls reach beyond American borders is gone. Here is where both sides converge on the same incoherence. The US lobbies to sell chips to China while building a coalition to reduce dependence on China. At the Beijing summit on May 14, Trump cleared roughly ten Chinese firms to buy up to 75,000 Nvidia H200 chips each ^[8]. Jensen Huang joined the delegation at Trump's last-minute insistence after seeing media coverage of his exclusion ^[9]. Trump's first request, as he put it:

We're not expecting any press releases, or any large declarations. It's just going to be purchase orders. — Jensen Huang

No chips shipped. China blocked them at the border ^[8]. The reason is substitution at scale: 560,000 Alibaba Zhenwu-series chips have been delivered to over 400 customers across 20 industries ^[10]. Nine domestic AI chips are now certified under China's national procurement framework, redirecting government and state-owned-enterprise demand to domestic alternatives like Huawei Ascend and Alibaba Zhenwu ^[11]. Nvidia's market share in China, in Jensen Huang's own words:

Nvidia's market share in China has effectively fallen to zero — Jensen Huang

The company whose China share was once 70% is now pitching Vera CPUs to Chinese clients, who plan to deploy them in overseas data centers to skirt controls ^[12]. The Chinese government, Huang acknowledged, now decides how much of its own market it wants to protect ^[10]. Meanwhile the US builds the other side of its dual strategy: critical-minerals pacts with India, Armenia, and Central Asian nations ^[13][14], an Apple-Intel domestic manufacturing partnership with a 10% government equity stake in Intel whose value grew from $100B to $600B in nine months ^[15], and the EU's Chips Act 2.0 aiming for 20% global semiconductor market share by 2030, restricting foreign cloud providers from public tenders in sensitive sectors ^[16]. All voluntary. All nascent. China mirrors the same contradiction from its side. It blocks Nvidia chips at the border while hosting Nvidia, Apple, and Intel at its Fourth International Supply Chain Expo in Beijing — 676 companies from 85 countries ^[17]. Its retaliation against the Pentagon blacklist targeted rare earth firms and defense contractors, not chips — consistent with its position as a net importer of advanced semiconductors ^[18]. A 22-article supply chain security framework took effect June 24, institutionalizing the power to investigate and block, but no known chip-specific investigation has yet used it ^[19]. The architecture is maturing into paper. The blocking, meanwhile, is already operational. The incoherence runs through the US system too. Anthropic warned that the US has 12 to 24 months to secure a decisive AI lead before China reaches near-parity by 2028, urging tighter controls and bigger enforcement budgets ^[20]. The same administration had just forced Anthropic to disable its models for foreign nationals to comply with export-control directives, cutting off major Indian IT firms ^[21]. The company pushing for stricter controls was punished by the controls regime. The company pushing to sell chips to China — Nvidia — was brokered into the presidential delegation. US officials still frame the competition in enforcement language — Bessent calls the US an "AI superpower" trailing nobody ^[22] — but the architecture underneath has moved from binding bilateral controls to voluntary multilateral diversification. The rhetoric persists while the coalition that made enforcement enforceable is gone. What's left is a system no one designed and no one can abandon. The US sells chips China won't buy and builds a coalition that can't enforce. China blocks chips it no longer needs and hosts the companies it just blocked. Both sides decouple selectively and integrate selectively, running dual strategies because neither works well enough to commit to alone.