DeepSeek, the Chinese startup that shook the world of technology this week with its groundbreaking artificial intelligence models, has taken a break for the lunar new year holiday.
The firm rocked the industry and global stock markets – wiping off around a trillion dollars in value from US stocks on Monday – by releasing cost-effective and open-source AI models rivalling the most advanced models of American tech giants OpenAI, Google and Meta.
And while the Silicon Valley, the Wall Street and the American political establishment, including new president Donald Trump, were still discussing the implications of DeepSeek’s breakthrough, the company’s headquarters in Hangzhou went quiet for the weeklong lunar new year holiday.
The last update to DeepSeek was issued late on Monday, and local media reports said the start-up’s offices were deserted for the holidays from Tuesday morning.
The lunar new year is based on a 12-year cycle, each linked to an animal in the Chinese zodiac paired with one of the five elements of metal, wood, water, fire, and earth. The festivities began on Wednesday, marking the year of the wood snake.
There was still buzz around the company, however, with security turning away uninvited guests at its headquarters, Tech in Asia reported.
The company claimed earlier this week that its advanced AI model, named R1, was trained at a fraction of the cost of its western rivals such as OpenAI’s ChatGPT, which employed a larger volume of expensive Nvidia chips to train.
DeepSeek said its AI model cost less than $6 million in computing power to train with older Nvidia H800 chips, in sharp contrast to the billions poured into creating similar systems by US companies like OpenAI, Meta and Google. The startup was forced to rely on older chips because the US had banned the export of advanced Nvidia systems to China in a bid to curtail its progress in AI.
DeepSeek’s success against such steep odds quickly led to sharp changes in the fortunes of major American AI companies. Shares of Nvidia plunged 17 per cent on Tuesday, marking the biggest single-day loss in market value of a company in history of over $500bn (£402bn).
Mr Trump warned the Chinese startup’s success was a “wake-up call” for the US tech industry.
“DeepSeek’s release of a premium level AI tool, available freely, with a reported miniscule development cost has shaken faith in Silicon Valley and American dominance in the rapidly developing AI market,” Richard Whittle, an economist at the University of Salford, said.
In spite of their remarkable achievement, the DeepSeek team keep a low profile. The startup’s controlling shareholder is Liang Wenfeng, who also co-founded a quantitative hedge fund called High-Flyer. It is unclear how much stake the hedge fund has in DeepSeek.
Government records indicate that High-Flyer owns patents for the chips used to train AI models.
Not much is publicly known about the other employees of Deepseek, whose official name, according to government records, is Hangzhou DeepSeek Artificial Intelligence Fundamental Technology Research Co Ltd.
A study posted by the company on its AI model in the arXiv database credits nearly 70 employees.
