U.S.–China AI Civilization Competition
Two civilizations training on each other's weights.
Kernel
By 2024 the U.S.-China AI competition had stopped pretending to be a normal trade dispute and started being treated as a civilizational contest by serious people in both countries. The American framing: liberal-democratic compute vs. authoritarian compute. The Chinese framing: a long-term technology sovereignty project that the AI moment merely accelerated. Both framings are partially correct; both produce policies that further entrench the bifurcation.
Origins
China's 2017 AI Plan declares 2030 dominance. The 2018 trade war begins decoupling. DeepMind's 2016 AlphaGo defeats Ke Jie in 2017 — read in China as a Sputnik moment. By 2023 China's open-model labs (Qwen, DeepSeek, ChatGLM, Yi) match Llama-class capability; by 2025 DeepSeek's R1-style training efficiency forces an honest American reassessment.
Doctrine
United States: maintain compute lead via export controls; lead at the frontier via the lab-NVIDIA-cloud triangle; bind allies into a single compute zone. China: substitute via efficiency; race up the open-weights curve; build a parallel chip industry; cultivate a domestic application stack indifferent to U.S. cloud.
Lineage
Each side's lineage is internal to its political culture. The American thread is Cold-War-flavored — Sputnik, Apollo, Star Wars, AI. The Chinese thread is sovereignty-flavored — opium war, semiconductors as the new fuel, indigenous innovation. The threads do not translate cleanly; this is part of the problem.
Conflicts
Tactical conflicts: export controls, chip smuggling, model distillation across borders, talent visas, university research restrictions. Strategic conflict: whose stack becomes the default infrastructure for the agent economy outside the OECD? The answer in much of Africa, Latin America, and Southeast Asia is increasingly Chinese.
Trajectory
The most likely 2030 outcome is bifurcation, not victory. Two parallel AI stacks, two parallel chip supply chains, two parallel agent payment rails, and a long unstable middle in the Global South. The Silicon Valley response — which is to assume Western models will dominate — is increasingly out of date.