AI is splitting along US–China lines
AI is the first major technology since semiconductors that the US and China are openly racing to control separately. Chip export controls, sovereign-AI initiatives, and dueling national strategies have made the AI stack a geopolitical fault line. The open question is whether the world ends up with one global AI ecosystem, two parallel ones, or something messier — and how much of the rest of the world has to pick a side.
Timeline
- October 7, 2022
The US Commerce Department imposes the first major chip export controls on China, blocking sales of advanced AI chips. The semiconductor industry calls it the most aggressive export-control action since the Cold War.
- October 17, 2023
The US tightens the rules to close loopholes Nvidia had used to keep selling slightly-below-threshold chips into China. The signal is that this is a long-term policy, not a one-off.
- May 22, 2024
DeepSeek releases V2, a Chinese model that matches GPT-4-class capability at a fraction of the cost. The launch breaks the assumption that the chip controls would freeze China's AI capability.
- January 20, 2025
DeepSeek-R1 releases, a reasoning model competitive with OpenAI's o1. Public reaction in the US ranges from "the controls failed" to "the controls created a more capable Chinese ecosystem."
- July 14, 2025
The Trump administration's second-term AI policy doubles down on chip controls and adds new restrictions on AI talent flow. Sovereign AI becomes the explicit US framing.
- November 30, 2025
Saudi Arabia, the UAE, India, and France each announce major sovereign-AI initiatives — partly hedges against being forced to pick a side, partly bets on owning their own stack.
- March 4, 2026
China announces export-control measures of its own on rare-earth minerals critical to advanced chip manufacturing. The decoupling becomes two-sided rather than one-sided.
Where things stand right now
The US–China AI split is hardening, not softening. Both sides have now imposed controls on the other; sovereign-AI initiatives in third countries are accelerating; and the rest of the world is increasingly being asked to pick whose stack to build on. A unified global AI ecosystem now looks less likely than it did even six months ago.