Chinese startup DeepSeek has released a “low cost” open-source artificial intelligence model rivalling OpenAI’s ChatGPT, drawing appreciation as well as concern from the Silicon Valley.
The R1 model made public last week appears to match OpenAI’s newer 01 models on several benchmarks. DeepSeek claims to have spent less than $6 million to train it compared to the hundreds of millions of dollars that American companies like Google, OpenAI and Meta have poured in to train their AI models.
R1 “outperforms other open-source models and achieves performance comparable to leading closed-source models”, the Chinese company says. “Despite its strong performance, it maintains economical training costs.”
DeepSeek’s new image-generation AI model, called Janus-Pro-7B and released on Monday, also seems to perform as well as or better than OpenAI’s DALL-E 3 on several benchmarks.
As word spread of the performance of the new Chinese AI model, stocks of leading tech firms in the US such as Nvidia and Oracle crashed on Monday, wiping out almost a trillion pounds in value off some of the world’s most prized companies.
Nvidia is the leading supplier of chips used to train and run AI models.
Share prices of American energy companies also suffered steep drops.
The reason: the new DeepSeek models seemingly belie the assertion by the Western tech ecosystem that developing advanced AI requires heavy investments of capital, electricity and water resources.
The Lawrence Berkeley National Laboratory, for one, predicts that advancements in AI will see data centers in the US consume as much as 12 per cent of the total electricity by 2028 as compared to 5 per cent in 2023.
DeepSeek, in contrast, offers the possibility of more cost-effective AI models that require much less capital and power to develop and run.
“The release of DeepSeek AI from a Chinese company should be a wakeup call for our industries that we need to be laser-focused on competing to win,” US president Donald Trump said.
“I view that as a positive because you’ll be doing that too, so you won’t be spending as much and you’ll get the same result, hopefully,” he said, adding that Washington will put more tariffs on foreign computer chips and semiconductors to return their production to the US.
DeepSeek is scripting success against heavy odds. The US, in an attempt to stall China’s progress in AI, has banned the export of advanced semiconductors and restricted the sales of Nvidia’s chips to the country.
The Chinese startup appears to have overcome this hurdle by refining its algorithms for efficiency and optimising the less sophisticated H800 chips.
Hancheng Cao, an assistant professor in information systems at Emory University, says the Chinese AI models are a “truly equalising breakthrough”.
They could be “great for researchers and developers with limited resources, especially those from the Global South,” he told MIT Technology Review.
The R1 mobile app has quickly climbed to the top of the Apple store’s free apps list, ahead of ChatGPT, sparking a debate on whether the Chinese startup posed a threat to its American competitors.
Alexandr Wang, head of software company Scale AI, based in San Francisco, says the success of R1 is a “wake-up call for America”.
“USA must out-innovate and race faster, as we have done in the entire history of AI, and tighten export controls on chips so that we can maintain future leads,” he said.
Some analysts, however, argue that DeepSeek’s success will be a shot in the arm for its American competitors due to the Chinese company’s approach of prioritising cost efficiency and open-source research.
“If training models get cheaper faster and easier, the demand for inference (the real world use of AI) will grow and accelerate even faster, which assures the supply of compute will be used,” Y Combinator chief Garry Tan said X.
Meta’s chief AI scientist, Yann LeCun, says the Chinese model’s success reflects on the “power of open research and open source”.
“People who see the performance of DeepSeek and think, ‘China is surpassing the US in AI’, you’re reading this wrong,” Mr LeCun said. “The correct reading is: ‘Open source models are surpassing proprietary ones.’”
Marc Andreessen describes Deepseek R1 “one of the most amazing and impressive breakthroughs”. “DeepSeek R1 is AI’s Sputnik moment,” the prominent venture capitalist said in a post on X.
DeepSeek was founded in 2023 by Liang Wenfeng, an alumnus of Zhejiang University, and incubated by High Flyer, a hedge fund he started in 2015.
The startup’s engineers reportedly consist of graduates from Chinese universities like Peking University and Tsinghua University.
“The emergence of China’s DeepSeek indicates that competition is intensifying, and although it may not pose a significant threat now, future competitors will evolve faster and challenge the established companies more quickly,” Saxo Markets chief investment strategist Charu Chanana told Bloomberg News.
