A speculative analysis argues that if AI capabilities plateau and become a commoditized utility, China's ability to rapidly scale energy infrastructure and produce cheaper tokens could allow it to dominate the global AI market, paralleling the offshoring of manufacturing.
The Premise: The Capability Plateau As a thought experiment, imagine a world where AI becomes good enough to fully automate the job of a senior software engineer, but right after that, the S-curve flattens. The returns on AI research start to diminish, and for the next 10 years, we are stuck with very slow improvements in the capability of frontier models. In that world, the rules of the AI arms race fundamentally shift. Frontier labs stop competing on capabilities and have to start competing entirely on price. Intelligence becomes a heavily commoditized utility. If that happens, I cannot see how China does not absolutely dominate the global AI market, because their "lag" behind US frontier labs (typically said to be 6-12 months) will become irrelevant. In a world of exponential growth, the 6 month gap means an ever increasing gap in capabilities in absolute terms. But on a flattening curve, it means almost nothing. If GPT-6 and Claude 5 are the absolute ceiling of AI, the difference between hitting that ceiling in January versus July is totally irrelevant over 10 years. On top of that, China can build and expand energy capacity at a speed the US simply cannot match. They don’t have the same issues with grid permitting, localized NIMBYism, or years-long environmental reviews. They can spin up gigawatts of nuclear or solar to power data centers by state decree. China can already produce tokens for way less than Western labs. When compute becomes a utility, this infrastructure gap will become fatal. We saw this exact movie in the late 20th century with physical manufacturing. The regulatory and labor arbitrage was an economic gravity that couldn't be defied, so the West offshored its physical production. If AI plateaus into a utility, we are looking at the offshoring of cognitive production. If the US wants to survive a commoditized AI market, it would require eradicating NIMBYism and deregulating energy grids at a speed our political system seems entirely incapable of. Curious to hear if anyone thinks the US has a viable way out of this if the models actually do plateau.
Anthropic released a paper arguing that the US and allies can lock in a 12-24 month AI lead by 2028 by restricting China's access to advanced compute and model outputs. It highlights China's use of loopholes, smuggled chips, and distillation to stay close, and frames advanced chips as the central bottleneck for AI power.
Explores the potential long-term risks if China becomes the global leader in AI, questioning whether cultural governance differences or technological dominance will shape future superpower behavior.
The article suggests that rapidly growing AI computing demand could make power supply a bottleneck. China, with its decades of energy infrastructure investments, has a head start in this competition, while the US may be neglecting energy issues due to its intense focus on chips.
Anthropic's policy paper outlines two scenarios for global AI leadership in 2028, depending on whether the US tightens export controls on compute to maintain its advantage over China.