China’s homegrown artificial intelligence models are being adopted quickly, but monetisation remains far behind leading US developers, according to new research from Rhodium Group. The firm estimates that all major Chinese AI models combined generate about 10% of the annual recurring revenue reported by OpenAI and Anthropic.
Among Chinese AI companies, DeepSeek had the lowest estimated ARR at $500 million, followed by MiniMax at $800 million and Moonshot at $1 billion. Z.ai told investors its latest ARR had reached $1.8 billion. Even adding ByteDance’s estimated $4 billion and Alibaba’s $2.4 billion, Chinese labs remain far behind Rhodium’s estimates of $40 billion for OpenAI and $65 billion for Anthropic.
The gap is even more striking when compared with valuations. Rhodium estimated valuation-to-revenue ratios of 50 times for Moonshot and 163 times for DeepSeek, versus 34 times for OpenAI and 21 times for Anthropic. The report said valuations relative to revenue “appear exorbitant” for Moonshot and DeepSeek. The disparity comes as Anthropic is reportedly expected to list in the US next month, OpenAI has pushed its IPO to next year, and Moonshot and DeepSeek are reportedly preparing potential listings.
Rhodium also highlighted open-source models as a key monetisation challenge for Chinese developers. While open-source access accelerates adoption, it makes capturing revenue from every use more difficult. “The financing gap means it will be far more difficult for Chinese frontier AI labs to scale sustainably,” said Logan Wright, partner at Rhodium Group.
On infrastructure, Rhodium estimated that more than 60% of equity investment in Chinese AI chips and servers came from state-affiliated sources. Chinese AI capital expenditure is projected to double this year to 932 billion yuan ($139 billion) and exceed 1.2 trillion yuan ($193 billion) in 2027, though the scale of China’s AI buildout remains only about 15%-20% of US investment. Chip shortages have constrained spending, but domestic supply from SMIC has increased, access to Nvidia H200 chips appears to be easing, and Chinese companies have been able to obtain advanced chips through third countries. Alibaba, Tencent and Baidu together spent 208 billion yuan on capital expenditure in the first half of 2026.
Separately, Anthropic and OpenAI are now pursuing smaller data centre deals of 20 to 30 megawatts, according to a CNBC report, alongside earlier multi-hundred-megawatt and gigawatt-scale commitments. Anthropic has sounded out smaller agreements in the UK and Nordics, while OpenAI has explored similar opportunities in the Nordics and US capacity. Anthropic signed a roughly $45 billion cloud deal with Nscale in August covering around 460 megawatts at a West Virginia site, and OpenAI has surpassed its original 10-gigawatt Stargate commitment.
The shift toward smaller sites reflects grid constraints and local opposition. Britain’s grid connection queue has grown from 41 gigawatts in late 2024 to roughly 125 gigawatts by mid-2025, with about 140 proposed data centre projects seeking 50 gigawatts. The Nordics offer average power prices below EUR 40 per megawatt-hour, compared with more than EUR 117 in the UK and over EUR 100 across Europe, making the region a faster deployment option.
The infrastructure race is unfolding amid growing AI safety concerns. King Charles III convened around 30 AI executives and policymakers at a private summit in Scotland, warning that AI could develop “darker capacities.” OpenAI recently disclosed six new cases of unexpected or concerning model behaviour, and security researchers reportedly used Anthropic’s Claude to breach OpenAI’s internal systems during a bug bounty exercise. OpenAI patched the flaw.
