How dependent is the U.S. on China for the AI buildout?

Two powerful secular investment themes are: 1) artificial intelligence (AI) and 2) rewiring of global trade. These two themes collide when it comes to the U.S. and China’s economic relationship. Broad AI adoption is crucial for both economies’ competitiveness – and the race to get there first remains fierce. Despite both sides’ efforts to “de-risk” their technology and critical supply chains from each other, these linkages are still very much there. Beyond trade, investors should not forget that the U.S.-China AI relationship also occurs via tech innovation itself, as China rapidly advances its own model development and application layers focused on open-source models and physical AI.

Over the past nine years, the U.S. administration and companies have focused on reducing direct trade ties to China. At surface level, this seems to have been achieved: China now represents only 9% of overall U.S. imports, down from 22% in 2017. However, beneath the surface, strong trade dependencies still exist¹:

• China’s economy is now geared towards high-tech manufacturing: Given muted domestic demand, Chinese policymakers have been focused on high-tech exports (which grew 71% at an annualized rate in July). Exports of AI-related inputs such as servers, computing hardware and electronic integrated circuits have seen the strongest growth.
• U.S. has reduced some of its high-tech imports from China: U.S. AI-related imports from China, such as computers/servers/storage and network equipment declined 70% and 41% year-over-year, respectively, last year. This occurred amidst U.S. targeted import bans under national security grounds.
• But strong dependencies on AI-related imports still exist: As illustrated in our “Navigating the trade policy fog” research, while imports of Chinese integrated circuits and semiconductors only represent 3% of all U.S. imports of this category, its imports of other AI-related inputs remain significant: network equipment at 18%, Chinese magnets at 29% and rare earth metals at 37%.

There are two ways investors can look at this dependency. Glass half-full: it provides a ceiling to rising tensions between the two countries (resulting in May’s “trade truce”). Glass half-empty: it injects downside risk to the AI buildout projections and accompanying productivity and profitability benefits. Worth monitoring: the U.S. administration’s potential new Chinese import bans (such as those of optical transceivers) and/or Chinese policymakers’ restrictions of key exports (such as rare earths last year).

Importantly, U.S.-China AI connections lie beyond trade. While U.S. model labs remain at the forefront today, Chinese homegrown models’ capabilities are rapidly catching up – and are cost-effective. Chinese-origin models’ share of tokens processed on OpenRouter² surpassed the U.S. this year at 60%. Even if broad enterprise AI adoption in the U.S. remains focused on U.S.-origin models, the competition from Chinese models can suppress prices – a new source of Chinese deflation hitting U.S. companies.

¹  Even leaving aside the transshipments of Chinese goods destined to the U.S. market via third countries.

² OpenRouter is a developer-facing service that routes API requests across hundreds of models, so its usage skews toward developer and agentic workloads. It under-represents enterprise workloads, which mostly run on the major cloud platforms or directly with a model developer, like OpenAI or Anthropic.

95006c78-37b0-11f0-90e0-19706d5121f1