When President Trump met Xi Jinping in Beijing on May 15, the semiconductor question that has defined US-China relations for half a decade was everywhere and nowhere. Xi reportedly framed Taiwan as a "red line" that, mishandled, would put the relationship "in jeopardy." Yet US Trade Representative Jamieson Greer said afterward that "there were no discussions of chip export controls" at all, and the Center for Strategic and International Studies concluded the summit revealed "how little progress has been made" on AI, export controls, and digital sovereignty. The two most powerful governments on earth spent a day together and left the central fact of the modern chip economy exactly as they found it: the advanced semiconductors both need are made in Taiwan, on Taiwan's terms.
Washington's leverage is slipping
The US theory of the case has been that export controls can slow China's access to frontier AI compute. That theory is under visible strain. In a January 14, 2026 proclamation issued under Section 232, the Trump administration reversed course and allowed Nvidia's H200 — and AMD's MI325X — to be sold into China, but attached a 25% tariff, the "cut" the government takes on each chip routed through the United States.
Then Beijing refused to play along. As CSIS notes, "no H200 chips have reportedly been shipped to China" despite Commerce issuing licenses, because the Chinese government is steering its firms toward domestic suppliers such as Huawei. The result is the worst of both worlds for Washington: it surrendered the clarity of a hard ban, booked little revenue, and handed Beijing a fresh incentive to accelerate the home-grown alternatives the controls were meant to prevent.
Taipei built its own chokepoint
If Washington's controls are leaking, Taipei's are tightening — and they are now law. Taiwan's Statute for Industrial Innovation, amended in May 2025, rewrote Article 22 to require government approval before companies invest in designated countries, regions, industries, or technologies, with the Ministry of Economic Affairs empowered to reject deals that threaten national security or economic development. A companion provision, Article 67-3, adds penalties for non-compliance. The official Science & Technology Law Institute describes the package plainly as protecting "critical technology sectors — particularly semiconductors."
Layered on top is the operational rule the market actually watches: an "N-minus" policy. Premier Cho Jung-tai confirmed that offshore fabs must run at least one generation behind Taiwan's leading edge (N-1), and by December 2025 officials including National Science and Technology Council Deputy Minister Lin Fa-cheng were describing a stricter N-2 posture — two full generations behind. In practice, TSMC's $100-billion-plus Arizona complex builds 4nm and 5nm today and 3nm-class chips by 2027, while 2nm, the A16 (1.6nm) node, and everything beyond stays in Taiwan, alongside most of TSMC's research workforce.
The case for the silicon shield
The logic deserves a fair hearing. Taiwan's concentration of leading-edge manufacturing — TSMC held roughly 71% of the global foundry market in the third quarter of 2025, per TrendForce — is widely understood as a "silicon shield": the world's dependence on Taiwanese fabs raises the cost of any Chinese coercion and gives democratic partners a stake in the island's security. Keeping the most advanced nodes and the engineers who develop them at home preserves that deterrent. From Washington's side, too, slowing Beijing's access to the compute that trains military AI is a legitimate aim, not a protectionist whim. Neither government is acting irrationally.
Where proportionality breaks down
But the evidence of the past eighteen months suggests both instincts are overshooting. Export controls that are broad and unpredictable have repeatedly produced the opposite of their intent: the H200 episode shows Beijing would rather forgo Nvidia silicon than depend on it, pouring the difference into Huawei and SMIC. Each escalation converts a commercial dependency — genuine US leverage — into a domestic Chinese industry.
Taiwan's N-minus rule carries a subtler risk. The single most effective way to blunt the "one earthquake, one blockade" fragility of the chip supply chain is geographic diversification — exactly what the Arizona fabs represent. A rule that deliberately keeps allied production two generations behind the frontier slows that diversification and guarantees that the highest-value, hardest-to-replace capacity stays in the one location everyone agrees is the flashpoint. The shield protects Taiwan; it does not make the global system more resilient, and at the margin it makes a disruption costlier for everyone else.
The chokepoint nobody fully controls
The honest conclusion of the Beijing summit is that no single capital commands the chip chokepoint. Washington can tax and license but cannot force China to buy or stop it from substituting. Beijing can refuse purchases but cannot manufacture 2nm at scale. Taipei can keep its crown jewels at home but depends on US security guarantees and Dutch lithography to make them.
Proportionate policy would lean into that interdependence rather than fight it: narrow, durable export rules aimed at genuinely military end-uses; predictable licensing that lets allied capacity scale; and an N-minus rule generous enough to let trusted partners build closer to the frontier. The frontier of computing is too important — and too concentrated — to be governed by reflexive control on three sides at once. The summit's real lesson is that the chokepoint will not be broken by tightening it further.