Last week, a staggering $1 trillion was wiped off global stock markets, triggered by the launch of an AI model from the Chinese start-up DeepSeek. The R1 model has made waves by challenging the assumption that the US dominates AI development. It can solve complex scientific problems and perform at a level on par with leading US AI models, yet it was developed at a fraction of the cost.
DeepSeek’s breakthrough has caused concern in Silicon Valley. R1 quickly became the most-downloaded free app on the US iOS App Store, displacing ChatGPT. The success raises a key question: Has China caught up in AI development just as the industry nears artificial general intelligence (AGI), where machines reach human-like intelligence?
DeepSeek, founded by billionaire Liang Wenfeng in 2023, has taken a Silicon Valley-inspired approach. It focuses purely on research without seeking external funding, allowing it to remain free from the pressures that often accompany state-backed companies. Liang’s approach reflects a boldness he believes is essential to innovation. “Silicon Valley is innovative because they dare to do things,” he said, noting that many in China lacked confidence in pioneering AI technology.
The company’s breakthrough came from its ability to train R1 using far fewer resources than US rivals. DeepSeek claims the final training for R1 cost only $5.6 million, compared to hundreds of millions spent by US firms like OpenAI. This achievement was, in part, made possible by China’s inability to access the most advanced US chips due to export restrictions. As a result, DeepSeek had to find innovative ways to make the most of less powerful technology.
Despite its success, the company has faced some controversy. OpenAI has claimed that DeepSeek used its own models in training R1, which would violate OpenAI’s terms of use, though such practices are believed to be common in the industry.
DeepSeek’s rise also challenges the narrative that AI requires massive amounts of capital and infrastructure to succeed. Some in Silicon Valley downplay its significance, suggesting that its innovations will be quickly replicated. However, others, like Sam Altman of OpenAI, acknowledged that R1 is “impressive” but noted that the rapid pace of AI development makes it difficult to determine whether DeepSeek will have lasting impact.
The real significance of DeepSeek’s rise lies in the way it is altering the balance of the global AI race. Its models are available at a significantly lower cost than those of US competitors, which has the potential to disrupt the AI market. For instance, while Meta charges $2.80 per million tokens, DeepSeek charges just 1.4 cents, making its technology much more accessible to developers worldwide.
In the end, DeepSeek’s success highlights the growing strength of China’s AI capabilities. While some argue that China is still behind in pushing AI frontiers, others believe that DeepSeek’s achievements are a sign that China is now a serious contender in the global AI race. This shift could have wide-ranging geopolitical consequences, with implications not just for technology but for military and global power dynamics as well.
