How small Chinese AI start-up DeepSeek shocked Silicon Valley

A small Chinese artificial intelligence lab stunned the world this week by revealing the technological recipe for its cutting-edge model, turning its reclusive leader into a national hero who has defied US efforts to curb China’s high-tech ambitions.

Dipsic, founded by hedge fund manager Liang Wenfeng, released its R1 model on Monday, explaining in a detailed paper how to build a large language model on a bootstrapped budget that can learn automatically and improve itself without human supervision.

US companies including OpenAI and Google DeepMind have made advances in reasoning models, a relatively new field of AI research that tries to match models to human cognitive abilities. In December, San Francisco-based OpenAI released Full version of o1 model But kept his methods secret.

Dipsic’s R1 release sparked a frenzied debate in Silicon Valley over whether better-resourced US AI companies, including Meta and Anthropic, could maintain their technological edge.

Meanwhile, Liang has become a focal point of national pride at home. This week, he was the only one A.I The leader was elected to attend a publicized meeting of entrepreneurs with Li Qiang, the country’s second most powerful leader. Entrepreneurs were told to “focus efforts on breaking out key core technologies.”

In 2021, Liang began buying thousands of Nvidia graphic processing units for his AI side project while running his quant trading fund High-Flyer. Industry insiders saw it as the whimsical act of a billionaire looking for a new hobby.

“When we first met him, he was this very nerdy guy with a terrible hairstyle talking about building a 10,000-chip cluster to train his own models. We didn’t take him seriously,” said one of Liang’s business partners.

“He couldn’t say his vision: I want to build this, and it will be a game changer. We thought it was only possible from giants like ByteDance and Alibaba,” the person added.

Liang’s position as an outsider to the AI ​​field was an unexpected source of strength. At High-Flyer, he made a fortune using AI and algorithms to identify patterns that could affect stock prices. His team excelled at using Nvidia chips to make money trading stocks. In 2023, he launched Dipsic, announcing his intention to develop human-level AI.

“Liang has built an exceptional infrastructure team that really understands how chips work,” says one founder of a rival LLM company. “He took Dipsy from the hedge fund with his best people.”

After Washington banned Nvidia from exporting its most powerful chips to China, local AI companies were forced to find innovative ways to maximize the computing power of a limited number of onshore chips—a problem Liang’s team already knew how to solve.

“DeepSeek’s engineers know how to unlock the potential of these GPUs, even if they are not state-of-the-art,” said an AI researcher close to the company.

Industry insiders say Dipsic’s singular focus on research makes it a dangerous competitor because it is willing to share its successes rather than protect them for commercial gain. DeepSeek has not raised outside funding or taken significant steps to monetize its models.

“DeepSick runs like the early days of DeepMind,” said an AI investor in Beijing. “It’s completely focused on research and engineering.”

Liang, who is personally involved in Dipsic’s research, uses the proceeds from his hedge fund trading to pay the highest salaries for the best AI talent. Along with TikTok-owner ByteDance, DeepSeek is known for offering the highest paid AI engineers in China, with workers based in offices in Hangzhou and Beijing.

“The offices at Dipsik feel like a university campus for serious researchers,” said the business partner. “The team believes in Liang’s vision: to show the world that Chinese people can be creative and create something from nothing.”

Dipsic and High-Flyer did not respond to requests for comment.

Liang styles Dipsic as a uniquely “local” company, staffed with PhDs from top Chinese schools, Peking, Tsinghua and Beihang universities, rather than specialists from US institutions.

In an interview with local media last year, he said his core team “doesn’t have foreign returnees. They are all local. . . . We have to develop our own top talent.” Dipsic’s identity as a purely Chinese LLM company has earned it appreciation at home.

Dipsic claims it used just 2,048 Nvidia H800s and $5.6 million to train a model with 671 billion parameters, a fraction of what OpenAI and Google spent to train models of comparable size.

Ritvik Gupta, an AI policy researcher at the University of California, Berkeley, said Dipsic’s recent model releases demonstrate that “there is no moat in AI capabilities”.

“The first person to train the model has to spend a lot of resources to get there,” he said. “But the second mover can get there cheaper and faster.”

Gupta added that China has a much larger talent pool of systems engineers than the U.S. who understand how to best use computing resources to train and run models more cheaply.

Industry insiders say that while Dipsic has shown impressive results with limited resources, it remains an open question whether it can be competitive as the industry evolves.

Returns to High-Flyer, its big backer, lagged in 2024, which a person close to Liang attributed to the founder’s focus being mostly on Dipsic.

Its US rivals are not stagnant. They’re building mega “clusters” of Nvidia’s next-generation Blackwell chips, generating computing power that threatens to once again create a performance gap with Chinese rivals.

This week, OpenAI said it was Creating a joint venture With Japan’s SoftBank, known as Stargate, planning to spend at least $100 billion on AI infrastructure in the US. Elon Musk’s xAI is massively expanding its Colossus supercomputer to contain more than 1 million GPUs to help train its Grok AI model.

Liang’s business partner said, “DeepSeek has one of the largest advanced computing clusters in China. “They have enough power now, but not for long.”

Additional reporting by Wenji Ding in Beijing