XPeng's Fuyao supercomputing hub hits 600 PFLOPS, cuts training from days to under an hour
bookwormengr · x · 2026-09-08
XPeng's full-stack AI development runs on the Fuyao supercomputing center in Ulanqab, Inner Mongolia, built with Alibaba Cloud: over 600 PFLOPS sustained (240 B200s at FP16), architected for large-scale multimodal training. Distributed infrastructure compressed core foundation-model training from multiple days to under an hour — a 170x+ iteration speedup. Having merged its automotive and robotics software divisions, XPeng feeds vehicle-collected visual data, spatial maps, and corner cases into Fuyao to train the universal VLA 2.0 model, a unified fleet data flywheel.
Related event: XPeng unveils in-house Turing chip and Fuyao supercomputing center(2 posts)→
More from Infra
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- US states halt new data centers as 72% of Americans oppose local builds — YvesMulkers · 2026-10-26
- $5M Seed Round Can't Afford One DeepSeek Training Run: GPU-less AI Startup's Three Paths — FinanceYF5 · 2026-10-26
- HBM demand shifts: low-stack inference plans shelved, Rubin to go 12-high only — BenBajarin · 2026-10-26
- CircuitHub unveils The Grid, a factory-scale robot making PCBs in the US twice as fast as Shenzhen — garrytan · 2026-10-26