U.S. vs. China · the Civilizational Race
Two AI stacks, two chip ecosystems, two definitions of victory.
Kernel
The U.S.-China AI competition is the dominant geopolitical fact of the 2020s technology cycle. By 2026 each side has its own frontier-lab portfolio, its own preferred chip supply chain, its own open-or-closed posture, and its own narrative about why it will win. The substantive technical convergence in 2024–2025 (DeepSeek R1, Qwen-Max, the Mistral-style efficiency frontier) has scrambled the comfortable U.S. assumption of permanent lead.
Frontline
Chip exports (the October 2022 controls + 2024 expansion). HBM and EUV equipment. Frontier-model release cadence. Open weights as a vector of soft power. Talent flows (Chinese-trained researchers in U.S. labs, U.S.-trained Chinese researchers returning, talent visa friction).
Doctrine — U.S.
Maintain the compute lead through export controls; ride NVIDIA-CUDA-cloud trinity at the frontier; bind allies into one compute zone; accept that AI is the defining national-security technology of the 21st century and act accordingly.
Doctrine — China
Substitute via efficiency. Race up the open-weights curve. Build an indigenous chip industry on whatever node generation is reachable. Cultivate a parallel application ecosystem (WeChat-class platforms, sovereign cloud, hardware vertical integration) that is structurally insulated from the U.S. stack.
Stakes
Whose stack becomes the default for the Global South in 2027–2030 may be the single most consequential question of the cycle. The U.S. assumption that English-language frontier models will dominate everywhere is increasingly questionable. In Africa, parts of Latin America, and most of Southeast Asia, Chinese open models are already common.
Outlook
The most likely 2030 outcome is a bifurcated world: two AI stacks, two chip supply chains, two agent-payment rails, and a long unstable middle. Neither side wins outright; both pay enormous compliance and inefficiency costs for the bifurcation; the global development frontier slows accordingly.