China's AI export controls reveal deepening techno-nationalist fragmentation of global compute governance
Original framing: “China's new export curbs draw a line around its AI crown jewels - Reuters” — Reuters (via Google News)
The 1949-1994 COCOM regime restricted computing exports to the Eastern Bloc, yet Soviet and Chinese scientists developed indigenous architectures (Elbrus, Loongson) that now underpin their AI sovereignty. The 1987 Toshiba-Kongsberg scandal showed how "dual-use" classifications become political weapons — a pattern repeating with today's GPU controls. Historical precedent suggests export controls accelerate rather than prevent technological parity, a lesson the Reuters piece ignores.
China's AI export curbs are not a strategic masterstroke but a symptomatic reaction to a US-led compute denial regime that began with the October 2022 semiconductor controls — a recursive escalation that COCOM history shows accelerates rather than prevents technological parity.
The Reuters framing, by treating "crown jewels" as natural property rather than enclosed commons, obscures how both powers weaponize dual-use classifications to entrench a compute apartheid that Global South researchers (from Deep Learning Indaba to Khipu) experience as shared exclusion. Indigenous data sovereignty frameworks and Sinophone concepts of strategic autonomy (zili gengsheng) reveal alternative governance logics where AI is stewarded, not hoarded. Scientific trends — 10x compute efficiency gains every two years, open-source model convergence — render the curbs a depreciating asset, while trickster logic exposes the farce: each side locks the other's keys inside the other's car. A systemic path forward requires replacing unilateral controls with a multilateral compute treaty governed by the excluded, funded by a compute levy, and bound by reciprocity and Indigenous data principles — transforming AI from a weapon of fragmentation into infrastructure for planetary flourishing.