Superpower Daily

Superpower Daily

Your reading space

Sign in
Huawei Sets 2027 AI Chip Roadmap Alongside New 4,000-Processor...Irregular Finds an AI Coding Agent Replaced Its Own Underlying...Snap Launches an AI Assistant Alongside Its $2,195 AR GlassesOpenAI’s Sponsored Agent Terms Leave Key User Safeguards Public...Stanford Publishes Paper2Agent to Turn Research Papers Into Wor...Amazon Receives Generac Warrants in Generator Deal, Sending Sha...Huawei Sets 2027 AI Chip Roadmap Alongside New 4,000-Processor...Irregular Finds an AI Coding Agent Replaced Its Own Underlying...Snap Launches an AI Assistant Alongside Its $2,195 AR GlassesOpenAI’s Sponsored Agent Terms Leave Key User Safeguards Public...Stanford Publishes Paper2Agent to Turn Research Papers Into Wor...Amazon Receives Generac Warrants in Generator Deal, Sending Sha...
Swipe

Rhodium Estimates Chinese AI Models Generate One-Tenth of U.S. Leaders’ Revenue

The estimates put a financial boundary around China’s low-cost model strategy, while raising harder questions about valuations and how labs capture revenue.

By 2 min read
Rhodium Estimates Chinese AI Models Generate One-Tenth of U.S. Leaders’ Revenue
Rhodium Estimates Chinese AI Models Generate One-Tenth of U.S. Leaders’ Revenue

Listen to this story

The audio brief

About 1:36
0:001:36
Read transcript
Chinese AI models are generating roughly one-tenth of the annual recurring revenue attributed to OpenAI and Anthropic, according to new estimates from Rhodium Group. That puts a financial boundary around China’s strategy of distributing relatively cheap, near-frontier models: broad use has not yet translated into comparable monetization. Rhodium estimates annual recurring revenue, or ARR, of about five hundred million dollars for DeepSeek, eight hundred million for MiniMax, one billion for Moonshot, four billion for ByteDance, and 2.4 billion for Alibaba. By comparison, it assigns forty billion dollars in ARR to OpenAI and sixty-five billion to Anthropic. ARR is a run rate: it annualizes a recent month of revenue. That makes it useful for tracking fast-growing businesses, but it is not the same as reported revenue for a completed year. That distinction matters here, because these are estimates rather than audited results. The numbers also make private valuations harder to ignore. Rhodium estimates Moonshot at roughly fifty times revenue, and DeepSeek at 163 times, compared with 34 times for OpenAI and 21 times for Anthropic. Those ratios reflect expectations as well as current business, but they raise the pressure to turn distribution into cash. There are signs the picture is changing. Z.ai says its ARR reached 1.8 billion dollars and forecasts three billion by year-end, though a forecast is not realized revenue. The key question is whether Chinese labs can capture more value through platform partnerships and revenue-sharing arrangements quickly enough to support their valuations.

Story brief

3 key points

Rhodium Group’s estimates suggest Chinese AI monetization remains far behind the leading U.S. labs, despite broad distribution of relatively cheap models. OpenAI and Anthropic are assigned $40 billion and $65 billion in ARR, while major Chinese players range from $500 million for DeepSeek to $4 billion for ByteDance. The estimates also imply stretched private valuations, especially for DeepSeek and Moonshot. The key...

  1. 01

    Rhodium estimates Chinese models collectively generate roughly 10% of OpenAI and Anthropic’s combined ARR.

  2. 02

    Estimated ARR: DeepSeek $500 million, MiniMax $800 million, Moonshot $1 billion, ByteDance $4 billion, and Alibaba $2.4 billion.

  3. 03

    Moonshot’s estimated valuation-to-revenue ratio is 50x; DeepSeek’s is 163x, versus 34x for OpenAI and 21x for Anthropic.

Rhodium Group’s new estimates put a stark number on the commercial gap in AI: Chinese models collectively generate about one-tenth of the annual recurring revenue attributed to OpenAI and Anthropic. The finding does not settle who will win adoption, but it sharpens a more immediate test: whether fast use of lower-cost models can turn into durable revenue.

The comparison uses annual recurring revenue, or ARR, a run-rate measure that annualizes a recent monthly revenue figure. It is useful for tracking rapidly growing companies, but it is not the same as reported full-year revenue.

A distribution strategy meets a revenue test

Rhodium’s earlier work characterized China’s approach as deploying and spreading cheap, near-frontier open-weight models, partly in pursuit of cost-efficient capability and technology sovereignty. The new revenue estimates expose the business-model pressure alongside that strategy.

At the company level, Rhodium estimates ARR of $500 million for DeepSeek, $800 million for MiniMax and $1 billion for Moonshot. Its estimates are higher for ByteDance at $4 billion and Alibaba at $2.4 billion, yet still well below the figures it assigns to the two U.S. leaders: $40 billion for OpenAI and $65 billion for Anthropic.

The central revenue gap
About 10%Chinese AI models combined

Rhodium estimates Chinese AI models collectively generate about one-tenth of the revenue attributed to OpenAI and Anthropic.

Valuations are running ahead of the estimates

The report’s concern is not simply that Chinese labs are smaller. It argues that some private-market valuations look unusually high relative to the revenue estimates. Rhodium puts Moonshot’s valuation-to-revenue ratio at 50 times and DeepSeek’s at 163 times, versus 34 times for OpenAI and 21 times for Anthropic.

Those ratios are estimates, not a verdict on any company’s future. But they make growth in monetization more consequential: a valuation can be supported by expectations, while revenue has to be captured from customers and partners.

The next move is taking a larger share

There are signs the revenue picture is still moving. Z.ai told investors its latest ARR was $1.8 billion and raised its year-end forecast to $3 billion from $2.4 billion. A forecast, however, is not a completed result.

Rhodium says Chinese labs are exploring ways to take more revenue from third-party providers that offer access to their models. It also estimates that state-affiliated sources supplied more than 60% of equity investment in Chinese AI chips and servers. The unresolved issue is whether broader distribution and new revenue-sharing arrangements can close the gap quickly enough to justify the valuations now attached to leading labs.

Sources

  1. rhg.comFighting Hubris in AI Strategy: A Layer-by-Layer Dissection of AI Tech Stack Competition
  2. cnbc.comOpenAI and Anthropic are making 10 times more revenue than all Chinese AI models combined, research group Rhodium says

Loading discussion...

Rhodium Estimates Chinese AI Models Generate One-Tenth of U.S. Leaders’ Revenue | Superpower Daily