Ok this is going to be a long one. For all our sakes, let’s stop obsessing over individual chip specs and upgrade to systems-level thinking

1/15 🧵

Part I: The Facts
The CloudMatrix 384 superpod draws ~560 kW IT power. Huawei has sold 750 *= ~420 MW. cnstock.com/commonDetail/7… x.com/ChrisRMcGuire/… 2/15
That's not enough for this mystery AIDC alone, but Huawei doesn't just sell pods. It sells nodes and cards.

Further, Huawei is not the only domestic player. Alibaba T-HEAD, Kunlunxin, MetaX, Biren, Moore Threads, and Enflame have all deployed tens of thousands of cards into individual clusters.machineyearning.io/p/chinas-silic…

export controls tried to kill China's chip designers... but may have accelerated them instead

evidence from the vanguard shows how memory breakthroughs, sparse computing, and energy-compute optimization are building decisive advantages for China's sovereign AI ecosystem

we're crossing an inflection point where their chips aren't just "good enough..." the Chinese AI stack may actually soon be ready for export

🧵↓ 1/ first, a recap. Initially for preventing dissemination of WMDs, the U.S. Entity List restricts access to TSMC process nodes, top-tier HBM, and EUV/DUV tools

all are bottlenecks, but HBM is particularly critical for high-performance AI workloads

semianalysis.com/2025/09/08/hua…

Silicon Valley is still scale-pilled - yeeting more GPUs, more data, more capital and hoping AGI falls out

it's increasingly clear that China is playing a different ballgame

we’re calling it Energy-Compute Theory. here's how it’s different, its implications for the new economy, and whether the US is falling behind

🧵↓ 1/ energy-compute theory (能算理论) reframes AI development as a physics problem: maximize your valuable intelligence per energy unit, subject to a minimum IQ (quality) constraint

that could be clearing an academic benchmark, a defect rate on an assembly line, etc