Been thinking about the recent GLM 5.2 news and the open weights angle everyone's running with but they're missing out on a completely different angle here.

Everyone's focused on the fact that a Chinese lab hit frontier-level performance and open-sourced it but the part worth sitting with is how. ZAI and the rest of Chinese labs were cut off from Nvidia in early 2025 so presumably no H100s, no H200s directly to them since then.
They crossed $128B on a model trained on probably Chinese silicon that lands within a few points of the frontier.
The export controls were meant to slow China down. What they actually did was force the competition onto power instead of chips, the one input China has in surplus. 543 GW of capacity added in a single year, larger than India's entire grid.
Since 2021, China has added more new power than the entire US grid holds, while the US grid is already straining under AI demand.
You can ration silicon. You can't ration electricity. And when the constrained side keeps shipping open weights anyway, the whole industry inherits the upside.