--- id: "2026-11-12-epoch-huawei-horinger-data-center-compute" url: "https://postcutoff.com/e/2026-11-12-epoch-huawei-horinger-data-center-compute/" as_of: "2026-11-22T13:32:00+01:00" date: "2026-11-12" date_precision: day category: hardware-compute importance: 3 confidence: medium status: [Partly confirmed] sources: 1 editor: human editor human_review: null version: "2026-11-16" --- As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-11-12-epoch-huawei-horinger-data-center-compute/ # Epoch: Huawei Horinger ≈ 7% of Colossus 2 Full title: Epoch AI: Huawei's Horinger campus, among China's largest AI data centres, has about 81,000 H100-equivalents, roughly 7% of SpaceX's Colossus 2 Epoch AI (Ben Cottier, Lynette Bye, Nov 12, 2026) estimated from satellite imagery that Huawei's AI campus in Horinger, Inner Mongolia runs about 105,000 Ascend 910C and 8,000 910B chips, about 81,000 H100-equivalents (90% CI 36k-141k) on ~286 MW, roughly 7% of Colossus 2. A new building due in early 2028 would raise it to ~141,000 H100e, while Colossus 2 is expected to reach 1.8M+. ## Key facts - Method: satellite images of Aug 18, 2026; cooling towers and chillers counted to model heat rejection, then IT power and chip counts, cross-checked with chip-sales estimates (four-month lag) and Huawei disclosures; operation inferred from water vapour and moving fans - Campus: three zones; six buildings, four with dense cooling (~286 MW IT); the North zone has few cooling towers and probably no powerful AI chips - Compute: ~81,000 H100e (90% CI 36,000-141,000); Huawei's figures put a 910C at ~0.76 H100e vs ~2.5 for Nvidia B200/B300 - Comparison: Colossus 2 (Tennessee, SpaceXAI) has ~14x Horinger's compute and should reach at least 1.8M H100e by 2028; Huawei's Wuhu campus is projected at 151,000 H100e by December 2027 - Epoch's wider finding: US AI data centres get about 2.5x more AI compute per unit of power than Chinese ones; China's compute constraints 'are unlikely to ease in the next two years' - Techmeme ran the analysis on Nov 16 ## What happened Epoch published a condensed version of a longer satellite-based report on one of China's largest AI campuses. Even counting all four cooled buildings as fully populated with Ascend chips, the site holds a small fraction of the compute of the largest US cluster. ## Why it matters It is one of the few independent, physical estimates of Chinese AI compute, published as US export controls tighten after the Nvidia scandal and as Chinese labs report capacity limits (`2026-11-15-deepseek-chat-limits-capacity`). ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 165 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 135 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 226 days after its cutoff - Grok 4.7 (training cutoff May 2026): 165 days after its cutoff ## Sources 1. [Epoch AI: What Huawei's Horinger data center reveals about China's compute gap (Nov 12)](https://epochai.substack.com/p/what-huaweis-horinger-data-center) (epochai.substack.com, official) ## Changes - 2026-11-16 (filed): Created (news agent) ## Related - 2026-10-27: [Epoch AI's chip-imports data](https://postcutoff.com/e/2026-10-27-epoch-ai-chip-imports-us-72-percent/index.md)