The Beijing start-up Naive AI is valued at over one billion dollars after just seven months – a record pace even for the Chinese AI scene. Investors like Tencent and IDG Capital have invested a total of 400 million dollars in three rounds into the company, which has not yet released a product of its own, as reported by the trade publication The Information.
Investors Pay Record Price for an Incomplete Product
Naive AI’s valuation skyrocketed within a few months. In April 2026, it was still around 800 million dollars after about 300 million dollars in capital, and by September, it reached 1.42 billion dollars after a total of 400 million dollars from three funding rounds – independently unverified, as Naive AI itself has not confirmed any figures so far. In addition to Tencent and IDG Capital, the MPCi fund and HSG, formerly known as Sequoia Capital China, also invested.
The technical approach for which the capital flows is unusual: Naive AI does not train its own base model from scratch but takes an existing open Chinese language model and refines it through retraining, fine-tuning, and reinforcement learning. The first own model is expected to be released in September 2026 as a freely downloadable open-weight model that users can customize without licensing costs – a bet on open rather than proprietary models, previously pursued by River AI, founded by Igor Babuschkin, with 1.1 billion dollars in capital. Fewer than 100 employees are currently working on the project.
Founder Dai Jifeng Brings an Unresolved Dispute
Behind Naive AI is Dai Jifeng, a private lecturer at Tsinghua University and co-developer of the image processing model InternVL. Before returning to research, he led teams at SenseTime and Microsoft Research Asia. Most recently, he worked until January 2026 as a technical advisor at MiroMind, the AI start-up of Shanda founder Chen Tianqiao.
The transition was not smooth: Dai stated in April to the media that MiroMind had tried to pressure him into moving abroad, which triggered his departure. MiroMind denied this and stated that its technology and intellectual property remained fully owned and had not been licensed to any third parties, as reported by implicator.ai. A public agreement between both parties is not yet known. Similar transitions of Chinese AI researchers from established labs to their own start-ups have recently attracted increasing attention.
Chinese AI Valuations Far Outpace Revenue
Naive AI enters a Chinese market already occupied by heavyweights like DeepSeek, Moonshot AI, and Z.ai – alongside established giants Alibaba and Baidu. According to estimates from the analysis firm Rhodium, Chinese AI models together generate only about ten percent of the annual revenues of OpenAI and Anthropic, while their valuations are significantly higher relative to revenue: DeepSeek is said to be valued at around 163 times its annual revenue, Moonshot AI at 50 times – compared to 34 times for OpenAI and 21 times for Anthropic.
Moonshot AI had previously made headlines with plans for an IPO in Hong Kong at a valuation of over 30 billion dollars. Naive AI shows that even a significantly smaller, still productless provider benefits from this influx of capital. There is no separate launch date for Germany or the EU for the planned open-weight model: as a freely accessible model, it is expected to be available for download worldwide simultaneously upon release.
It will be crucial whether Naive AI can demonstrate with the first model announced for September that targeted retraining of foreign architectures pays off economically – or whether the valuation is solely based on investors’ hopes of having gotten in early with a potential winner. The record rise occurs during a phase in which Chinese providers increasingly decouple their valuations from actual revenues.
