Geopolitics, Export Controls and US–China Competition

AI is the first general-purpose technology whose key input is made in a handful of places that governments can count. The most advanced AI chips are designed mostly in California, fabricated mostly by one company in Taiwan, packaged with memory made mostly in South Korea, and printed on machines made only in the Netherlands. That concentration turned chips into an instrument of state power. From October 2022 the United States used export controls to cap the computing power China could buy and the chipmaking tools it could import, and it widened those rules in October 2023, December 2024 and January 2025. Then policy turned. In 2025 the Trump administration scrapped the global "AI diffusion" rule, banned and then re-licensed Nvidia's China chip in exchange for a share of revenue, and in December 2025 offered the more powerful H200 to China for a 25% cut. Beijing answered with rare-earth controls, an antitrust case against Nvidia and instructions to its own companies to buy Chinese chips instead. Meanwhile Huawei and SMIC kept shipping chips the rules were meant to prevent, smugglers moved servers through Southeast Asia, Chinese labs such as DeepSeek showed that strong models could be trained on restricted hardware, and the Gulf states bought their way into the top tier of compute. This section explains the chokepoints, the rules and how they changed, China's response, the smuggling problem, Taiwan risk, the Gulf deals, and how an investor, founder or policymaker should read the next move.