China’s open model shock has reached silicon valley
- The AI trade was built on scarcity. Open-weight Chinese models are starting to make that scarcity look less like a moat and more like a toll booth.
- The real disruption is not just performance. It is control, privacy and cost. Enterprises may not need the best model; they may only need one that is good enough and lives on their side of the wall.
- If token costs keep falling while open models improve, the market’s assumption of endless pricing power for the US model leaders becomes harder to defend.
- The capex boom can still run, but the value may migrate away from the closed-model gatekeepers and toward chips, infrastructure, enterprise integration and whoever can make AI cheaper to own.
China’s open model shock
This is a risk we flagged back in mid-June, when GLM 5.2 looked less like another Chinese benchmark challenger and more like a heavy swing at the valuation scaffolding beneath the US AI trade. The point was never that Zhipu had suddenly built the undisputed best model in the world. The point was that intelligence was becoming cheaper, more portable and less geographically captive, just as the US hyperscaler complex was pouring hundreds of billions into an infrastructure buildout priced on the assumption that frontier capability would remain scarce, differentiated and richly monetizable.
A few weeks later, that risk is moving from the research notes into the real market conversation. Silicon Valley has spent the past two years selling investors a simple story: the best models would remain scarce, expensive and tightly controlled by a small group of US firms sitting behind high walls of compute, talent and proprietary data. GLM 5.2 is beginning to test whether those walls are a fortress or simply a very expensive toll booth.
Zhipu’s model arrived at an awkward moment. Washington briefly tightened access around Anthropic’s leading systems, then appeared to reverse course, just as developers and companies were looking for cheaper, more controllable alternatives. The timing could hardly have been worse for the incumbents. The market was reminded that the product it depends on is not entirely in its own hands, while a credible substitute was waiting just outside the gate...
Author
Stephen Innes
SPI Asset Management
With more than 25 years of experience, Stephen has a deep-seated knowledge of G10 and Asian currency markets as well as precious metal and oil markets.
