Six months ago, Chinese AI was still a catch-up story. Not anymore. Six of the top ten most-used AI models on global developer platforms are now Chinese. The most capable openly licensed model in the world was released this month under an MIT license with no restrictions. Output tokens from DeepSeek cost 29 times less than Claude Opus 4.8. And the US government action designed to contain this shift may have accelerated it.

This is not a story about geopolitics, or benchmarks, or one remarkable quarter. It is a structural repricing of where AI value accrues, and it has direct consequences for every portfolio with exposure to the AI theme, on either side of the trade.

In this report, we map the full competitive landscape: the Chinese LLM companies already listed and the four unicorns heading toward IPO between Q4 2026 and Q4 2027. We also make a direct call on those two stocks: where the risk/reward is attractive, where it isn't, and what the earnings catalysts are.

The conclusion is not comfortable for consensus AI positioning. Intelligence is becoming infrastructure, the way bandwidth became infrastructure after 2005. The returns are migrating up the stack. The companies that own the distribution layer, the proprietary data, and the applications will capture them. Those whose valuations rest on sustained AI pricing power face a structural headwind that is already in motion.

The large-cap Chinese tech stocks have consolidated alongside their US hyperscaler counterparts, the macro AI trade is already well-owned at that end of the market. The more interesting opportunity lies further out on the risk curve, in a segment that is only now becoming accessible to public market investors. A small number of entry points exist today. That window is unlikely to stay open for long.

The first publicly traded vehicles to express this view directly now exist. This report tells you exactly what they are, what they're worth, and how to think about sizing them.

Chinese Internet stocks (KWEB ETF) declined sharply