China’s STAR50 slumps 7.7% as Meta’s spare AI compute stokes overcapacity fears

The STAR50’s sharp drop is less about fundamentals changing overnight and more about how jittery public markets have become around AI‑infra narratives. Meta’s plan to sell surplus AI compute was enough to spook investors into wondering whether the industry is over‑building GPU capacity. Chinese analysts quoted by 36Kr push back, framing Meta’s idle GPUs as a company‑specific mismatch between infra build‑out and model commercialization, not the end of the AI capex cycle.

For the AGI race, the underlying reality is that demand for compute is still growing faster than supply, especially for training‑class systems. But sentiment swings like this matter because they influence how fast capital markets will keep funding the build‑out of data centers, power infrastructure and chip factories. If investors start to believe that AI infra is a bubble rather than a long‑term secular trend, financing costs rise and timelines can slip.

The episode also shows how tightly coupled global AI stories have become: a Meta decision in the US ricochets through Chinese growth indices within hours. That interdependence makes it harder for any one country to decouple its AI industrial policy from global market psychology.