Satellite analysis: Huawei's Horinger campus has just 7% of Colossus 2's AI compute
Epoch AI · rss · 2026-11-13
Epoch AI used commercial satellite imagery and public sources to estimate the AI compute at Huawei's Horinger, Inner Mongolia data center, as a lens on the US-China compute gap. The North zone's sparse cooling equipment suggests it lacks power-dense AI servers; the four dense-cooling buildings total 286 MW of IT power, three already operational.
Assuming a four-month sale-to-deployment lag, they estimate 105,000 Ascend 910C and 8,000 910B chips are operational, equal to 81,000 H100-equivalents (90% CI: 36k–141k). For context, the 910C performs at 0.76 H100e while Nvidia's B200/B300 reach 2.5 H100e. Horinger is among China's largest AI data centers (only VNET Bayin Ulanqab and Huawei Wuhu likely rival it), yet holds just 7% of xAI's Colossus 2 capacity — about 14x smaller.
Outlook: a new South-zone building is expected operational by February 2028, growing campus compute over 70% to 141,000 H100e, but Colossus 2 is projected to reach at least 1.8 million H100e by 2028. Still, 10% of Horinger's compute (8,100 H100e) could train a DeepSeek-V4-Pro-scale model in a month; fully combined, a GPT-6 Astra-scale model would take about a year.
More from Infra
- Japan inference startup ai& deploys heterogeneous chips, plans 100MW+ data centers — DavidBennett__ · 2026-11-13
- Accord GPU open beta queues ComfyUI, Blender, Ollama to stop VRAM fights — MrAddams_LibraLogic · 2026-11-13
- Ex-Nvidia CUDA engineer: 'CUDA is not the moat — Nvidia's only moat is its people' — AlpinDale · 2026-11-13
- AMD expands PsiQuantum partnership with planned equity investment in photonic quantum computing — sven_ai · 2026-11-13
- Nvidia downgrades hit seven as analysts say depths 'not fully probed' — SumitGup · 2026-11-13
- AMD Advancing AI World Tour Japan set for Dec 1 in Tokyo with KDDI, Microsoft, AMD execs — DavidBennett__ · 2026-11-13