Wall Street and Silicon Valley got clobbered on Monday over rising fears about DeepSeek a Chinese artificial intelligence startup that claims to have developed an advanced model at a fraction of the cost of its US counterparts.
Shares of Nvidia and other major tech giants shed more than $1 trillion in market value as investors parsed details.
DeepSeek claims it built its AI model in a matter of months for just $6 million, upending expectations in an industry that has forecast hundreds of billions of dollars in spending on the scarce computer chips that are required to train and operate the technology.
By comparison, OpenAI CEO Sam Altman haspublicly stated that his firms GPT-4 model cost more than $100 million to train. Last year, Dario Amodei, CEO of rival firm Anthropic, said modelscurrently in development could cost $1 billion to train and suggested that number could hit $100 billion within just a few years.
DeepSeek immediately surged to the top of the charts in Apples App Store over the weekend displacing OpenAIs ChatGPT and other competitors. The rapid rise has sparked panic that the US could lose its AI advantage to China.
Billionaire tech investor Marc Andreessen called DeepSeeks model AIs Sputnik moment a reference to the Soviet Unions launch of an Earth-orbiting satellite in 1957 that stunned the US and sparked the space race between the two superpowers.
Here is what to know about DeepSeek:
DeepSeek was founded in May 2023. Based in Hangzhou, China, the company develops open-source AI models, which means they are readily accessible to the public and any developer can use it.
DeepSeek released details earlier this month on R1, the reasoning model that underpins its chatbot. The AI firm turned heads in Silicon Valley with a research paper explaining how it built the model.
DeepSeek says the model excels at problem-solving despite being much cheaper to train and run than its rivals.
DeepSeek claims that the performance of its R1 model is on par with the latest release from OpenAI.
Alexandr Wang, the CEO of Scale AI, recently told CNBC that DeepSeek is the top-performing, or roughly on par with the best American models.
DeepSeeks top shareholder is Liang Wenfeng, who runs the $8 billion Chinese hedge fund High-Flyer. High-Flyer has an office in the same building as its headquarters, according to Chinese corporate records obtained by Reuters.
The AI chatbot has already faced allegations of rampant censorship in line with the Chinese Communist Party’s preferences.
DeepSeek admitted that its “programming and knowledge base are designed to follow Chinas laws and regulations, as well as socialist core values,” according to an output posted on the US House’s select committee on China.
The chatbot self-censored its responses when asked about China’s leader Xi Jinping, the Tiananmen Square massacre of 1989, China’s human rights abuses toward Uighurs in its Xinjiang Province and whether Taiwan is a country, according to examples circulating on social media.
DeepSeek is a privately held startup and is not publicly traded in the US.
Shares of Nvidia plunged a whopping 17% in Monday trading on panic related to DeepSeek, erasing more than $600 billion in value from its market cap. However, the long-term threat that DeepSeeks success poses to Nvidias business model remains to be seen.
For example, analysts at Citi said access to advanced computer chips, such as those made by Nvidia, will remain a key barrier to entry in the AI market. Wedbush analyst Dan Ives described the chaos around DeepSeeks launch as a buying opportunity.
Nvidia is also heavily involved in President Trumps $500 billion Stargate AI infrastructure project.
Nvidia bulls have noted that DeepSeek still used its chips to train the model, albeit with a less costly footprint than previously thought possible.