Trump Lifts Ban on Nvidia H200, Sparking 2-Million-Chip Chinese Buying Spree and Emergency TSMC Ramp-Up

Great-power technology controls have flipped before: after the 1949 establishment of CoCom, the U.S. barred supercomputer exports to the USSR—until détente in the early 1970s briefly loosened rules, only to tighten again after the 1979 invasion of Afghanistan. Today’s H200 waiver rhymes with that episode: a short-term commercial opening that may hasten a rival’s capacity. The move spotlights two structural trends: 1) semiconductor chokepoints as instruments of statecraft, and 2) the growing power of corporate supply chains that can pressure governments when billions are on the table. Whether the U.S. tax share and licensing can truly monitor downstream use recalls the 1987 Toshiba-Kongsberg scandal, where restricted milling machines ended up in Soviet submarines despite paperwork. On a 100-year horizon, the question is not who ships which GPU in 2026 but whether export controls can meaningfully slow diffusion of compute know-how; history suggests technical knowledge eventually leaks while bans often accelerate indigenous efforts—China’s own Ascend and Broadcom-co-designed parts may mature faster because of this very pivot. If AI becomes as foundational as electricity, today’s waiver is either a pragmatic revenue grab or an inflection that cedes strategic advantage; its true weight will only be clear decades after H200s are obsolete scrap metal.