Ben Thompson argues in Stratechery that AI depends on a precarious chip supply

Ben Thompson argues in Stratechery that AI depends on a precarious chip supply

Ben Thompson’s article was published on February 25, 2025. Thompson argues that AI’s promise rests on an unstable arrangement: U.S. AI development depends heavily on chips made by TSMC in Taiwan, while chip bans may reduce China’s dependence on Taiwan. He says that dependence has helped deter conflict, so weakening it could leave Taiwan more exposed even as the U.S. remains reliant on its chips. Thompson rejects the idea that TSMC could simply take over Intel’s foundries: their manufacturing processes differ, and moving equipment alone would not transfer TSMC’s know-how or remove its reliance on Taiwan. He proposes restoring Chinese access to TSMC-made chips and leading Nvidia chips, while restricting China’s access to semiconductor manufacturing equipment. He also calls for U.S. subsidies for trailing-edge chip production and a series of interventions to build leading-edge capacity at Intel Foundry. These proposals, he argues, involve serious risks and trade-offs; he does not present them as easy fixes. His conclusion is that AI’s opportunity is also precarious, and that making Taiwan more important to China is the price of buying time for U.S. chip capacity to develop. Whether his proposed policy mix could strengthen that deterrent without increasing other risks remains unknown.

Publication: 1f5aacae-33bc-48dd-482b-bd9027e4d5a0

attribution — pass: The reading attributes the article to Ben Thompson at Stratechery, which matches the source. It frames his diagnosis and proposals as his arguments throughout.

I believe this plan, with all of the risks and sacrifices it entails, is the best way

summary — uncertain: The chip-policy thesis, the Intel takeover critique, the proposals and the conclusion are faithful, and the title and question fit. However, central terms are left unexplained for a general reader and no definitions are supplied: TSMC as a contract chipmaker, foundry, leading-edge, and trailing-edge. Trailing-edge chips (older, cheaper chips used in cars and appliances) carry Thompson's vulnerability argument. Smallest repair: add plain one-clause glosses or term-library tooltips for these terms.

There are a lot of chips in the world — both on AI servers and also in everything from cars to stereos to refrigerators — that don’t need to be on the cutting edge

date — pass: February 25, 2025 matches the dateline and the UTC day of originalPublished. The limitations note the later retrieval date. The later listing entries at the end of the page and the Anthropic update note are not used. The plans are reported as Thompson's proposals, not as outcomes, and there is no hindsight framing.

Tuesday, February 25, 2025

support — pass: Each claim matches its source excerpt. The deterrence premise, the point that a TSMC takeover would not solve the Taiwan problem, the subsidy proposals and the price-in-the-meantime conclusion are all source-stated. The closing unknown is framed as open rather than asserted.

And, as long as China needs TSMC chips, they have a powerful incentive to leave Taiwan alone.

media — pass: No media is selected and no verified candidates exist, so there is nothing to misidentify.