Michael Darius Eastwood · Independent AI alignment researcher
Published
Chapter 8 of Infinite Architects is titled The Chokepoint. It is the chapter policy readers have quoted most in the six months since the book came out. This note summarises what it argues, in the book's own terms, without adding claims the chapter did not make.
The chapter identifies the concentration of advanced semiconductor manufacturing: TSMC and Samsung as leading fabs, ASML as the sole supplier of EUV lithography, a small number of design houses producing frontier AI accelerators. It notes that the concentration is not accidental; it results from decades of accumulated expertise that cannot be relocated on the timescale on which AI capability is scaling.
Any jurisdiction that hosts the concentration can require conditions of manufacture as a matter of domestic law. Because there is no substitute supplier on the relevant timescale, the market follows. The chapter argues this is the mechanical basis on which international AI governance can, unusually, be effective: you do not need every government on Earth to agree. You need Taiwan, South Korea, the United States, the Netherlands, and perhaps a few others.
Three. First, a treaty; the chapter names it the HARI Treaty (Hard-Alignment, Recursive-Improvement) and analogises it to nuclear non-proliferation, noting that the AI challenge is in some ways easier because the chokepoint is narrower and more visible. Second, a certification; the chapter names it the Eden Mark, drawing an analogy with Fair Trade, and proposes that chip designs be submitted to an International AI Ethics Authority for architectural review before manufacture. Third, the meltdown-trigger mechanism at the chip layer as a fallback.
The chapter is explicit that the chokepoint is temporary. "The technological lead that makes the chokepoint effective may last five years. It could shift. Market share may shift. China's domestic capabilities will advance." The framework must be established while the chokepoint exists, "knowing that the chokepoint will eventually close." That is a genuinely rare feature of policy writing on this topic: the author names the expiry date of their own leverage.
The architecture may be circumvented; design assumptions may prove wrong; meltdown triggers depend on computational assumptions future systems may transcend; the chokepoint may close through domestic capability development, alternative manufacturing techniques, or the emergence of quantum computing paths that do not require conventional semiconductors. The chapter treats each of these as real risks rather than rhetorical concessions.
It does not propose the exact statutory language for any of the mechanisms. It does not name the enforcement authority in the treaty structure. It does not settle the criterion for review; whether the design includes caretaker doping at the required level and whether meltdown triggers are properly integrated is a criterion frame, not a criterion. Those are engineering choices the chapter identifies as urgent without pretending to have finished them.
A structural argument about supply-chain concentration as leverage, three mechanism families, an explicit expiry date on the leverage, and a candid list of limits. The chapter has aged unusually well in the six months since publication because it has been quoted in policy vehicles it did not name and inverted in policy vehicles it did not anticipate. That is the shape of an argument doing real work.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.