Chinese AI Start-Up Achieves $1bn Valuation In Eight Months

An AI start-up founded by computer scientist Kai-Fu Lee has attained a valuation of greater than $1 billion (£810m) in less than eight months based on an open source artificial intelligence model that has beaten top US rivals including Meta’s Llama 2 in key metrics.

The latest funding round for 01.AI, with participants including Alibaba Group Holding’s cloud unit, valued the firm at over $1bn, Lee said, without disclosing exact details of the amount raised.

Lee, chief executive of venture firm Sinovation Ventures, began bringing together the team for 01.AI in March and started operations in June, and is also serving as chief executive of the start-up.

Known for founding Microsoft Research Asia more than 25 years ago and for his best-selling books AI Superpowers: China, Silicon Valley and the New World Order and AI 2041: Ten Visions for Our Future, Lee is a lifelong AI enthusiast who has already backed several AI “unicorns” at Sinovation.

Performance metrics

Beijing-based 01.AI is currently making its open source model Yi-34B – named after the 34 billion parameters used in its training – available to developers around the world in Chinese and English, with plans to add further languages later on.

Over the weekend AI evaluation firm Hugging Face posted metrics ranking Yi-34B first for whoat are called pre-trained base LLMs (large language models), beating Meta’s Llama 2 amongst others.

He said offering Yi-34B as an open source model capable of running on relatively inexpensive systems was a way of making the technology widely available to researchers who might not need or be able to afford the largest, most expensive LLMs.

The company is planning a six billion parameter 6B model, but also one with 100 billion parameters and a proprietary model to serve as the basis for a range of commercial products, Lee said in multiple interviews published over the weekend.

AI chip stockpile

He said the company plans to debut its first app within the calendar year and hinted at concepts related to productivity and social media.

Chinese firms have been restricted from buying the high-end US GPUs needed for AI training since October 2022, with further restrictions introduced last month, but Lee said the company had stockpiled all the chips it might need for the next 12 to 18 months before the latest round of export controls took effect.

“With a very high-quality infrastructure team, for every 1,000 GPUs, we might be able to squeeze 2,000 GPUs workload out of them,” he told TechCrunch.

He said 01.AI now has a staff of more than 100 people, about half of whom are LLM experts from multinational tech firms, including Google, as well as researchers from major Chinese companies.

Investors have been pouring money into AI since OpenAI launched ChatGPT a year ago, with Elon Musk’s xAI most recently launching a chatbot called Grok over the weekend.