Taiwan as chokepoint
The world’s leading-edge chips sit on an island whose gas lasts less than two weeks. Coal lasts about seven. Allied fabs take years. Seven blockades have not browned the island out.
TSMC makes about 90 percent of the world’s most advanced chips. The island that hosts those fabs, according to Pat Gelsinger, the former Intel chief, holds less than three weeks of energy reserves.
He was citing a Wall Street Journal piece from early July. A later fetch of that Journal / CSIS grain splits the tank. LNG lasts less than two weeks. Coal lasts about seven. Coal already sits above the three-week falsifier. LNG sits below it. The brownout clock is the gas, not the pile of coal. A blockade, Gelsinger said, does not need a shot fired. After the gas goes, the island browns out. A fab that loses power does not come back for 90 days. That restart figure is still single-sourced.
The island of Taiwan has less than three weeks. … Less than three weeks of energy reserves.
Pat Gelsinger, citing the Wall Street Journal, July 2026
Bloomberg Economics puts a Taiwan conflict at about $10 trillion globally — about 10 percent of world GDP, worse than Covid. That is a war-and-seizure number. The same shop’s blockade year-one print is smaller and still ugly: Taiwan minus 12.5 percent, China minus 8.9, the United States minus 3.2, the world minus 5.3. It is not the Great Depression, and it is not the $10 trillion war. The cheaper tool is the one Gelsinger is pointing at: close the strait, wait, and let the lights go out.
Walter Russell Mead, on a September Invest Like the Best tape, named a still cheaper face of that tool. Xi is “committed to trying to get it,” he said, without an invasion alarm-clock. The path he describes is a soft blockade: “any ship that wants to go to Taiwan first has to stop at a Chinese port and go through a customs clearing process to assert sovereignty.” Japan, in the same telling, “would see Chinese control of Taiwan as a direct threat to Japanese independence” and is raising defense spending. Coercion without an amphibious landing. It does not flip the reshoring chain, and it is not a new ticker.
Seven times, and the lights stayed on
China has closed the Taiwan Strait seven times in the last four years, Gelsinger said. He hedged the count — “I think.” The exercises are real enough that he called the capability demonstrated, not theoretical. “This isn’t a theory.”
The lights stayed on. Seven blockades, zero brownouts. Whatever those operations were, they stayed below the threshold that would turn a three-week reserve into a 90-day fab outage. The base rate of blockade-to-brownout is 0 for 7. That is the strongest fact in the record, and no one Gelsinger was talking to addressed it.
He also has a life’s argument to protect. Gelsinger staked his tenure on American fabs. “Taiwan is fragile, therefore build in Arizona” is the conclusion he is motivated to reach. The reserve figure is checkable. The blockade count is checkable. Until someone fetches the Journal piece, the three-week clock is second-hand.
The fabs that actually got built
Allied reshoring is the insurance policy. It takes three to five years. The chokepoint is now.
When Gelsinger came back to Intel in 2021, the United States was building about 12 percent of leading-edge silicon. He puts that number at about 18 percent today. In the same breath he said the TSMC-to-Intel wafer ratio went from 5-to-1 to 7-to-1. Both can be true only if the onshoring accrued to TSMC’s Arizona lines, not to Intel. “TSMC factories are up and operating at scale,” he said. Intel “is starting to be a real foundry.”
What Intel has published as foundry proof is performance per watt, not a die-yield percentage. A June 3 note, written before the VLSI symposium, said 18A yields were above 60 percent and rising 7 to 8 percent a month. Official T1.2 and a June 16 newsroom post published 18A-P PPA — more than 9 percent at iso-power, or more than 18 percent lower power — and Power Boost. They did not publish a die-yield percent. Treat the 60 percent figure as a pre-symposium or other-channel claim. Risk production of 18A-P was dated June 2026. Second-quarter Foundry revenue was $5.8 billion; external foundry was $293 million, still mostly internal.
C.C. Wei, TSMC’s chief executive, announced an additional $100 billion for Arizona — more leading-edge wafer fabs and, for the first time on US soil, advanced-packaging plants. He would not give a schedule. In the same call he said TSMC is also building 13 leading-edge and advanced-packaging fabs in Taiwan. The absolute US number rises. Whether Taiwan’s share falls is not established. Overseas-fab gross-margin dilution is 2 to 3 percent early and widens to 3 to 4 percent later. The penalty for allied capacity is real, and it grows with the mix.
US share of leading-edge
Gelsinger’s figures. The TSMC-to-Intel wafer ratio widened over the same stretch, from 5:1 to 7:1.
Chamath Palihapitiya has a shorter clock. Eighteen months, he said in May, and Taiwan stops being “an important moment of conversation,” because US fabs will be one or two nanometers from doing what Taiwan does. Western intelligence puts the PLA’s Taiwan-scenario readiness at the 2027 centenary of the army. December 2025 drills were described as the largest Taiwan-focused exercises ever. One clock is receding. The other is approaching. Neither has resolved the other.
Michael Kratsios, the administration’s science adviser, put a different lock on the same island. China, he said in late August, has been trying to figure out EUV lithography since the 2019 export controls. “No breakthrough has happened. There’s nothing more important to their economy than solving that scientific problem and haven’t been able to do that.” That is independent speaker color on the ASML chokepoint the wiki already carries. It is not a new step, and it is not a reason to reprice a reactor name.