US Stocks Plunge $1 Trillion Amid DeepSeek's AI Breakthrough
The Chinese artificial intelligence startup has since reported experiencing “large-scale malicious attacks” that have disrupted new user registrations on its platform.
China-based artificial intelligence (AI) startup DeepSeek is raising concerns about American AI dominance after releasing its AI model, R1, on January 20, reports the Financial Times.
Just one week after the release of the DeepSeek model, US tech stocks tumbled, with Nvidia’s stock dropping 17 percent, wiping out US$589 billion in market capitalization, making it the biggest drop in a single day for a US company. The Nasdaq Composite and S&P 500 also saw decreases of 3 percent and 1.5 percent, respectively.
The Chinese AI startup has since reported experiencing “large-scale malicious attacks” that have led the company to limit new user registrations on its platform.
Established by Chinese entrepreneur and businessman Liang Wenfeng in the eastern Chinese city of Hangzhou in 2023, DeepSeek reportedly spent just $5.6 million training its R1 model, compared to the estimated $100 million to $1 billion that Silicon Valley companies typically spend per model.
Wenfeng, who used funds from his hedge fund High-Flyer to buy Nvidia chips, attracted top AI talent from China with competitive salaries. The company claims that it used only 2,000 of Nvidia’s reduced capability H800 chips—far fewer than those used by US developers—and optimized their efficiency by skipping steps that US firms typically prioritize.
Despite limited resources, the R1 model has performed well in several third-party tests, with DeepSeek itself claiming that its R1 model outperforms OpenAI’s o1-mini model in several benchmarks. DeepSeek’s alternative has gone on to be adopted by companies seeking a more cost-effective solution, according to the Wall Street Journal.
OpenAI, initially founded as a non-profit to advance artificial general intelligence (AGI) for the benefit of humanity, transitioned to a "capped-profit" model in 2019, allowing investors to earn limited returns on their funding. This shift culminated in OpenAI becoming a privately held company in 2024.
Critics have argued that the move has distanced the organization from its original mission of ensuring open and equitable access to AI advancements. DeepSeek, on the other hand, is tuned to OpenAI's original mission, NVIDIA Senior Research Manager and Lead of Embodied AI (GEAR Lab) Jim Fan noted in a post on X.
We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely.
— Jim Fan (@DrJimFan) January 20, 2025
DeepSeek-R1 not only open-sources a barrage of models but… pic.twitter.com/M7eZnEmCOY
US President Donald Trump has said that DeepSeek should serve as a “wake-up call” for US companies, emphasizing the need for increased innovation and competitiveness in the AI sector, while OpenAI CEO Sam Altman lauded DeepSeek for coming up as "a new competitor."
deepseek's r1 is an impressive model, particularly around what they're able to deliver for the price.
— Sam Altman (@sama) January 28, 2025
we will obviously deliver much better models and also it's legit invigorating to have a new competitor! we will pull up some releases.
Meanwhile, venture capitalist and advisor to US President Donald Trump Marc Andreessen referred to the launch of DeepSeek's AI model as "AI's Sputnik moment," drawing a parallel to the 1957 Soviet satellite launch that spurred the US space program.
Deepseek R1 is AI's Sputnik moment.
— Marc Andreessen 🇺🇸 (@pmarca) January 26, 2025
These developments come amid what many have called an escalating "AI arms race" between the US and China, which has seen US sanctions prohibit the export of Nvidia’s most advanced H100 chips to China since September 2022. In response, Nvidia created the less powerful H800 chips for the Chinese market, but these too were banned from export to China in October 2023. DeepSeek had reportedly stockpiled Nvidia A100 chips before the sanctions came into effect, in addition to using the reduced capability H800 chips to train its models.
