DeepSeek and China’s other leading AI companies together generate only around one-tenth of the combined revenue of OpenAI and Anthropic, Nikkei Asia reported on 10 November, citing estimates from the US research nonprofit Epoch AI: an annualized $11 billion for six Chinese firms against $105 billion for the two US labs.
Anthropic’s own filing widens the gap. Its S-1/A says run-rate revenue was approximately $81 billion as of 31 October, up $10 billion from 30 September. That one-month increase is about 91% of the six Chinese firms’ combined $11 billion, and $81 billion is about 7.4 times it, by our arithmetic; the estimates come from different months and are not like-for-like.
Business Pill · Open Weights Explained
An evergreen Business Engineer explainer on open weights: rent a model through an API or run it yourself. It teaches the general idea only and does not discuss Epoch’s estimates or these companies.
The key insight: As we read it, the gap is a pricingstory as much as a demand story: Epoch AI estimates China’s six leading AI firms at about $11 billion of annualized revenue, while Anthropic’s run rate rose $10 billion in October alone, and Epoch ties the Chinese shortfall to open weights that let other hosts serve the same models.
Epoch AI’s analysis, published on 6 October, puts annualized revenue at $65 billion for Anthropic and $40 billion for OpenAI, against $4.0 billion for ByteDance, $2.4 billion for Alibaba, $1.8 billion for Z.ai, $1.0 billion each for Moonshot and DeepSeek and $0.8 billion for MiniMax. Epoch says its revenue estimates are from different months ranging from June to September.
The ByteDance and Alibaba figures are specific to Model-as-a-Service revenue, Epoch says. ByteDance alone is about 36% of the Chinese total, and the four model-focused firms add up to $4.6 billion, by our arithmetic. Nikkei notes that The Information reported in September that DeepSeek’s annualized revenue run rate had doubled to reach $1 billion.
Annualized revenue: Anthropic’s run-rate of about $81 billion at 31 October and the implied $71 billion at 30 September (its S-1/A), OpenAI’s $40 billion and China’s six leading AI firms’ combined $11.0 billion (Epoch AI estimates from June to September), and Anthropic’s $10 billion October increase. The 30 September figure and the 91% comparison are our arithmetic; the estimates come from different months.China’s six leading AI firms by annualized revenue, from Epoch AI’s estimates (run rates reported June to September; ByteDance and Alibaba are Model-as-a-Service only). ByteDance’s 36% share is our arithmetic.
Why Open Weights Cap the Price
Epoch’s argument is that Chinese labs release their model weights, which increases adoption “but comes at the expense of their pricing power”. As of 16 September, only about 25% of all GLM-5 tokens over the previous 24 hours were served through Z.ai’s own API, it says, and in the last 24 hours only 28.4% of DeepSeek V4 Pro’s tokens on OpenRouter were served by DeepSeek.
The sharpest case is Z.ai’s GLM 5.3 Flash. Epoch says Z.ai’s share of the model’s OpenRouter tokens fell from 88% to 22% over the 20 days after it released the weights, by which time Z.ai was competing with 26 rivals. By contrast, Epoch says a closed-weight model such as Anthropic’s Fable is almost always provided at the same list price of $10 per million input tokens.
Nikkei reports that open models account for more than 70% of token usage on OpenRouter, and that DeepSeek’s V4.1-Flash-1017 model costs roughly 5%-8% of what the American companies charge. UBS analyst Wei Xiong estimates gross profit margins for the Chinese companies in the 20%-40% range, Nikkei says.
Where Anthropic’s $81 Billion Comes From
Anthropic defines run-rate revenue as consumption-based fees in the trailing 28 days multiplied by 13, plus monthly subscription fees multiplied by 12, and says it “is not a substitute for revenue prepared in accordance with GAAP or a forecast of revenue”. We set out the October figure when the amended filing landed.
Nikkei adds that Anthropic’s revenue totaled $33.2 billion in the nine months through September, 13 times the year-earlier level, and that the company could raise $100 billion in its IPO. On reported revenue rather than run rates, we earlier compared Anthropic with two listed Chinese labs, Zhipu and MiniMax.
China’s Labs Look for Capital
DeepSeek and Moonshot AI are looking to raise money from investors and potentially go public, Nikkei reports. Z.ai and MiniMax listed in Hong Kong in January, and Nikkei says shares in both have slumped around 80%-90% from their peaks; both reported net losses for the six months through June.
The Chinese labs are also adding revenue streams, according to Nikkei: Z.ai has signed a contract with Amazon to offer its GLM-5.3 model through Amazon’s cloud, and Moonshot AI charges licensing fees for large-scale commercial deployment of Kimi K3. Our recent China coverage includes ByteDance’s Doubao token volumes and Alibaba’s Qwen4-Max preview.
The Structural Read
As we read Epoch’s data, releasing weights trades revenue for reach: Z.ai’s share of its own GLM 5.3 Flash tokens on OpenRouter fell from 88% to 22% in 20 days, while a closed model such as Anthropic’s Fable is almost always provided at the same list price.
As we read the two sets of figures together, the comparison is unusually lopsided rather than precise: Epoch’s Chinese estimates run from June to September, while Anthropic’s $81 billion is a 31 October run rate that the company itself says is not a forecast.
Epoch AI, How do Chinese AI companies make money?
“Overall, releasing model weights leaves Chinese labs competing with third-party hosts to serve their own models.”
Three Implications
FOR PRICING Nikkei reports that DeepSeek’s V4.1-Flash-1017 costs roughly 5%-8% of what the American companies charge, and that third parties offered DeepSeek’s V4 Pro for 87% less than the official price as of 16 September.
FOR MARGINS Epoch puts Z.ai’s API and developer-platform gross margin at 24.6% in the first half of 2026, against an 85% API-serving gross margin for Anthropic’s Opus 4.8 by mid-2026; UBS’s Wei Xiong estimates 20%-40% for the Chinese companies, Nikkei says.
FOR CAPITAL Nikkei reports that Anthropic could raise $100 billion in its IPO, while DeepSeek and Moonshot AI are looking to raise money and Z.ai and MiniMax shares have slumped around 80%-90% from their peaks.
The framework opens: “Every technology stack has layers. Some are scarce, some are abundant.” Its rule is to close the scarce layer and open the abundant one.
As we read it, Epoch’s figures show the cost of opening the model layer itself: Chinese labs gain adoption but share the serving revenue with third-party hosts, while Anthropic keeps its frontier models closed. The framework predates Epoch’s analysis and does not discuss these companies.
What Is Not Established
Epoch’s figures are estimates assembled from run rates reported in different months, and the ByteDance and Alibaba numbers cover Model-as-a-Service revenue only, so the $11 billion total is not a measure of all AI-related revenue at those groups. The comparison with Anthropic’s 31 October run rate mixes a filed company metric with third-party estimates from earlier months.
We did not contact Epoch AI, Nikkei, Anthropic or the Chinese companies. The 91%, 7.4 times, 36% and $4.6 billion figures are our arithmetic.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Epoch AI estimates China’s six leading AI firms at about $11 billion of annualized revenue, roughly a tenth of OpenAI and Anthropic, Nikkei reports; Anthropic’s filing shows its run rate rose $10 billion in October alone, to about $81 billion.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. We read Nikkei Asia’s article of 10 November 2026 in full, Epoch AI’s analysis of 6 October 2026, which is the source of the company estimates, and Anthropic’s S-1/A of 10 November 2026 for its run-rate figure and definition. We did not contact Epoch AI, Nikkei, Anthropic or the Chinese companies.
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.
