Max Bessler: divided AI market blunts US leverage

Bessler builds his argument by comparing AI to other markets that split into commodity and premium halves. In medicine, he notes, 90% of prescriptions are generic drugs but only 12% of spending, while brands take 88% of spending on 10% of prescriptions. The same pattern, he argues, is appearing in AI: Vercel's September 2026 report showed open-weight models processing 56% of August's gateway tokens at 14% of spend, and CNBC found the share of tokens U.S. companies used on Chinese models through OpenRouter above 30% since February, reaching as high as 46%.

He reads that shift against the way Washington has turned past technology waves into leverage. Huawei's global smartphone share collapsed from 20% in 2020 to virtually nothing outside mainland China by 2024 after Washington cut it off from Google's Android in 2019 and TSMC's chips in 2020. Cisco's routers once carried most of the world's online data, a position that later became a means of control. Edward Fishman's chokepoint test — dominant concentrated share, no short-term substitutes, asymmetric pain — is, in Bessler's view, failed by open models that are free to download.

The historical parallel he leans on most is Japan's takeover of the DRAM market, where NEC, Toshiba, and Hitachi pushed Japan's share to 80 percent by 1987, yet Japan gained little coercive capability. Bessler suggests the U.S. may have to accept that the model layer divides regardless of policy, and that the levers worth keeping are narrow chip controls and American cloud hosting — owning where the model runs rather than which model is used.