China’s artificial intelligence models are being used at enormous scale, but that surging adoption still hasn’t translated into anything close to comparable revenue against their American rivals, according to new estimates published Thursday by research firm Rhodium Group. All of China’s AI models combined generate roughly $10.7 billion in annual recurring revenue, about 10 percent of what OpenAI and Anthropic report between them, a gap that raises real questions about whether current valuations attached to several leading Chinese AI startups can be justified by their actual commercial performance.
The scale of the disparity becomes clearer once you break down the individual figures. OpenAI’s annual recurring revenue, or ARR, an industry metric that annualizes a recent monthly revenue figure by multiplying it by twelve, stands at roughly $40 billion, while Anthropic’s reaches approximately $65 billion. On the Chinese side, DeepSeek posted the lowest ARR among major players at just $500 million, followed by MiniMax at $800 million and Moonshot at $1 billion. Z.ai told investors this week that its latest ARR had climbed to $1.8 billion. Even the two largest Chinese tech conglomerates dwarfed those AI-specific figures without coming close to matching their American counterparts, with ByteDance reporting $4 billion and Alibaba $2.4 billion, both still a fraction of OpenAI’s revenue alone.
What makes this gap genuinely striking rather than simply a predictable byproduct of China’s AI industry being younger is how the valuations attached to these companies compare. Rhodium’s analysis found that Moonshot currently carries a valuation-to-revenue multiple of roughly 50 times, while DeepSeek’s implied multiple reaches an extraordinary 163 times. Both figures dwarf the comparable multiples for the leading American labs, 34 times for OpenAI and 21 times for Anthropic. Rhodium’s report described the situation plainly, stating that valuations relative to revenue appear exorbitant for Moonshot and DeepSeek at present, a conclusion that lands with particular weight given how closely investors watch exactly this kind of metric when assessing whether a startup’s private valuation reflects genuine business fundamentals or simply speculative enthusiasm.
The timing of this analysis intersects directly with a wave of pending public listings across the Chinese AI sector. Moonshot has reportedly filed confidentially for a Hong Kong listing, though the company has said it does not comment on market rumors or speculation regarding any such filing, and DeepSeek is separately reported to be preparing its own public offering. On the American side, Anthropic is reportedly expected to list on a US exchange next month, while OpenAI has pushed its own IPO plans back to next year. That parallel timing means public markets could soon be asked to price Chinese and American AI companies against genuinely comparable disclosure standards for the first time, a test that, based on Rhodium’s current figures, could prove uncomfortable for at least some of the Chinese names heading toward a listing with revenue multiples this far out of line with their American peers.
Part of the explanation for China’s weaker revenue capture traces back to a fundamental structural difference in how the two countries’ AI industries are built. Rhodium’s report noted that Chinese AI labs are actively searching for better ways to secure revenue as third parties access their underlying models to build and offer their own services, a monetization challenge rooted in how openly Chinese AI models are typically released. Because Chinese models are strongly open in nature, anyone with sufficient computing hardware can download and run them directly without ever routing usage, or payment, through the original developer. That stands in sharp contrast to how most leading American AI companies operate, keeping their most capable models closed and monetizing access directly through paid subscriptions and API usage, a structure that captures revenue far more reliably even when actual usage volumes are comparable.
The revenue concentration data adds another layer to the picture. Payments platform Ramp found that the top 1 percent of customers supply roughly 80 percent of revenue at both OpenAI and Anthropic, illustrating just how much of even the leading American labs’ commercial success currently depends on a relatively small number of large enterprise customers rather than broad-based, evenly distributed adoption. That concentration cuts both ways for interpreting China’s weaker figures, since it suggests American AI revenue itself remains somewhat fragile and dependent on retaining a narrow set of major accounts, even as it currently outpaces China’s by a wide margin.
None of this has slowed China’s underlying investment appetite, however. AI capital spending in China is set to roughly double this year to 932 billion yuan, approximately $139 billion, with projections pointing toward surpassing $193 billion by 2027. That buildout still runs at only about 15 to 20 percent of comparable US investment levels according to Rhodium’s broader assessment, and Rhodium separately noted that more than 60 percent of equity investment flowing into Chinese AI chips and servers has come from state-affiliated sources, with the research firm suggesting government funding has proven more useful supporting the hardware side of China’s compute buildout than it’s likely to be in directly subsidizing frontier AI labs themselves going forward. Cloud AI services currently make up only around 13 percent of revenue at Chinese hyperscalers and telecom operators, another indicator of how much monetization upside remains theoretically available even as current figures lag so far behind American competitors.
It’s worth noting the limits of Rhodium’s own analysis, which draws on the latest available figures gathered through this past summer, a period during which usage of Chinese AI models has reportedly surged from considerably lower levels earlier in the year. That timing gap means the true current revenue picture for companies like Moonshot and Z.ai could already look somewhat different than what this specific snapshot captures, particularly given how quickly adoption metrics have been moving across the sector in recent months. Whether that faster adoption curve eventually closes the revenue gap with OpenAI and Anthropic, or whether China’s more open-model business strategy structurally caps how much revenue these companies can ever realistically capture regardless of usage volume, remains the central question hanging over the sector as several of its biggest names head toward public market scrutiny in the months ahead.
For more coverage of the global AI industry and technology market analysis, visit Business Tech.
