Epoch AI estimates in a new report published on 6 November that Huawei’s AI data center in Horinger, Inner Mongolia, has about 81,000 H100-equivalents of computing capacity, just 7% of the world’s largest AI data center, SpaceX’s Colossus 2 in Tennessee. Epoch says Colossus 2 has “about 14 times as much computing capacity as Huawei Horinger”.
The report, by Ben Cottier, uses satellite imagery and public sources. Epoch puts the Horinger estimate’s 90% confidence interval at 36,000 to 141,000 H100-equivalents.
Business Pill · Concentration
A short explainer on concentration. It teaches the general idea only and does not discuss this report.
The key insight: As we read Epoch’s estimates, Huawei’s Horinger campus holds about 81,000 H100-equivalents, 7% of Colossus 2, and on Epoch’s own 2028 projections Colossus 2 would still have about 12.8 times its capacity, by our arithmetic.
Epoch says it bought high-resolution satellite images of the three zones and estimated power from rooftop cooling: “based on the type, size, and number of cooling units visible, we estimate their heat rejection capacity and, in turn, the buildings’ IT power capacity”.
It excluded the North zone, which “has far fewer cooling towers than the West and South zones”, leaving four buildings. “Our cooling model estimates a total of 286 megawatts (MW) of IT power capacity across these four buildings”, the report says, with other methods giving 304 MW from substation details and 230 MW from backup generator capacity. Epoch estimates three of the four are running and the fourth will become operational by December.
AI computing capacity in H100-equivalents, as estimated by Epoch AI in its report of 6 November 2026 from satellite imagery and public sources: Huawei Horinger today (90% CI 36,000 to 141,000), Horinger in early 2028, and Colossus 2’s expected capacity of at least 1.8 million by 2028. These are Epoch’s estimates, not company figures.
What Was Posted on X
Huaweiโs AI data center in Horinger, Inner Mongolia, is one of the largest in China by AI computing capacity. Yet it has just 7% as much computing capacity as the worldโs largest AI data center, Colossus 2 in Tennessee.
This capacity estimate is from our new report examining the Horinger facility and Huaweiโs broader AI infrastructure plans using satellite imagery and public sources.
Epoch AI announced its report on X, saying Huawei’s Horinger data center is one of the largest in China by AI computing capacity yet has 7% as much capacity as Colossus 2, based on satellite imagery and public sources.
The Chips Inside
Assuming no export-controlled chips are involved and a reported rack occupancy of about 76% for the Horinger area, Epoch writes that “about 105,000 Ascend 910C chips and 8,000 Ascend 910B chips are operational to date”. It estimates the Ascend 910C at about 0.76 H100-equivalents per chip, based on 8-bit performance, and says Nvidia’s B200 and B300 are about 2.5 times more energy-efficient.
Epoch also writes that Huawei “makes far fewer chips than Nvidia”, and that because of the efficiency gap, “a given amount of power capacity can only support 40% of the compute of leading US facilities”. Against Epoch’s chip sales estimate of 907,000 Ascend 910C chips sold by the end of March 2026, it says Horinger had used 11% of the total 910C stock by August.
The Gap in 2028
Epoch expects a new building in the South zone to be operational in February 2028, lifting the campus by over 70% to about 141,000 H100-equivalents (90% CI: 65,000 to 227,000). It says this “would still leave it at about the same share, or a smaller one, of Colossus 2’s capacity”, which it expects to reach at least 1.8 million H100-equivalents by 2028.
On those two projections, Colossus 2 would have about 12.8 times Horinger’s capacity in 2028, by our arithmetic, against about 14 times today. Epoch adds that Huawei’s sister campus in Wuhu is projected to reach 151,000 H100-equivalents by December 2027.
Huawei’s Wider Build-Out
Epoch says it has identified four sites owned by Huawei and four owned by China Telecom, China Mobile, China Unicom and the Shenzhen Longgang government that appear to host CloudMatrix384 supernodes, each with 384 Ascend 910C chips. It estimates 12 Huawei buildings are likely running Ascend 910C chips today, and says early renderings show at least 27 data center buildings planned at Horinger alone.
The report’s conclusion: “China appears to have no shortage of energy for data centers”, but US AI data centers “support 2.5 times more AI compute per unit of power”.
As we read the report, the method is physical: rooftop cooling equipment counted from satellite images, cross-checked against substation and backup-generator capacity.
As we read the figures, Horinger had used about 11% of the Ascend 910C stock Epoch estimates was sold by the end of March 2026, so the campus is large but not the whole of Huaweiโs deployment.
As we read Epochโs conclusion, power is not the limit in China; the report puts the constraint on the number and efficiency of chips.
Epoch AI, in its Horinger report
“China appears to have no shortage of energy for data centers.”
Three Implications
FOR COMPUTE COMPARISONS Epoch puts Horinger at 81,000 H100-equivalents, with a 90% confidence interval of 36,000 to 141,000.
FOR CHIP SUPPLY Epoch estimates about 105,000 Ascend 910C chips are operational at Horinger, out of 907,000 it estimates were sold by the end of March 2026.
FOR THE NEXT BUILD Epoch expects a new Horinger building in February 2028 to lift the campus by over 70%.
The framework says: “The structural bottleneck isn’t computing, it’s electrons.”
As we read it, Epoch’s Horinger report describes the mirror case: Epoch says China has ample energy, and the limit sits in the chips. The framework predates Epoch’s report and does not discuss it.
What Is Not Established
These are Epoch’s estimates from satellite imagery and public sources, not figures from Huawei; Epoch says some of its sources are withheld to protect their availability. It assumes no export-controlled Nvidia chips are present, describing their use as “possible but unlikely”, and that Horinger matches the area’s 76% rack occupancy. Epoch also notes that a campus’s compute is often split into separate clusters and cannot necessarily be combined to train one model. The 12.8 times figure is our arithmetic from Epoch’s 2028 projections.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Epoch estimates Huawei’s Horinger campus at about 81,000 H100-equivalents, 7% of Colossus 2, and says China has the energy but is held back by fewer and less efficient chips: each megawatt carries about 40% of the compute of leading US sites.
A note on sourcing. All capacity, chip and power figures are Epoch AI’s estimates from its report of 6 November 2026, based on satellite imagery and public sources; they are not figures from Huawei. The 12.8 times figure is our arithmetic from Epoch’s 2028 projections.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
