Faisal Bashir | Lightrocket | Getty Images<\/p>\n
BEIJING \u2014 China’s artificial intelligence models may be enjoying rapid adoption but that isn’t yet translating into revenue, raising questions about company valuations.<\/p>\n
All of China’s AI models combined generate only about 10% of the revenue reported for OpenAI and Anthropic, U.S.-based research firm Rhodium Group said in estimates published Thursday. That’s based on reports using an industry metric called annual recurring revenues, which attempts to capture fast growth by multiplying a recent monthly figure by 12.<\/p>\n
DeepSeek’s ARR was the lowest among major Chinese AI companies, at $500 million, the Rhodium report said. MiniMax was next at $800 million, followed by Moonshot at $1 billion.<\/p>\n
Z.ai told investors on Wednesday its latest ARR was $1.8 billion, according to a transcript seen by CNBC.<\/p>\n
But even with $4 billion for ByteDance and $2.4 billion for Alibaba, that remained far lower than the $40 billion generated by OpenAI alone, not to mention $65 billion for Anthropic, Rhodium said.<\/p>\n
Critically, the low revenue is far from keeping pace with how investors are valuing the Chinese startups.<\/p>\n
“Valuations<\/a> relative to revenue appear exorbitant for Moonshot and DeepSeek at present,” the Rhodium report said. The analysts noted estimated ratios of 50x and 163x, respectively, for the two startups.<\/p>\n That’s far above the 34x for OpenAI and 21x for Anthropic, the report said.<\/p>\n Z.ai forecast<\/p>\n To be sure, the Rhodium analysis could reference only the latest available figures from this summer, and usage of Chinese AI models has skyrocketed from low levels earlier this year. <\/p>\n Z.ai on Wednesday said it now expects its ARR by the end of the year to reach $3 billion, up from $2.4 billion previously forecast.<\/p>\n Rhodium also pointed out that Chinese AI labs are exploring ways to get a larger cut of revenue from third parties offering access to the models, whose open-source nature allows anyone with sufficiently capable hardware to download and run it independently of the developer.<\/p>\n U.S. models are mostly closed, and the cost per task<\/a> for leading AI models from OpenAI and Anthropic is far higher than that of Chinese models, according to AI-comparison firm Artificial Analysis.<\/p>\n “The financing gap<\/a> means it will be far more difficult for Chinese frontier AI labs to scale sustainably,” Logan Wright, partner at Rhodium Group, said in a statement to CNBC. He co-authored the report with research analyst Endeavour Tian.<\/p>\n “They will be heavily dependent upon a favorable climate in the equity market\u2014historically that’s not an easy bet in China,” he said. “Government funding has been helpful on the hardware side of the buildout of compute capacity, but similarly will probably balk at direct funding for the frontier labs.”<\/p>\n Rhodium estimated more than 60% of equity investment in Chinese AI chips and servers came from state-affiliated sources.<\/p>\n It’s been a volatile year for Chinese AI companies that have listed.<\/p>\n Z.ai shares rose more than 5% in Thursday morning trading, recovering from a decline earlier this week following news of its second major fundraise in two months<\/a>. The Hong Kong-traded stock has tumbled to levels seen this spring, after briefly more than tripling in price over the summer. <\/p>\n