A seemingly unknown model, DeepSeek R1, entered the fiercely competitive AI space on Jan. 10. Its launch immediately sparked shock and surprise throughout the global tech community, as the model quickly demonstrated performance on par with established frontier models, including OpenAI’s flagship o1 reasoning model, the most advanced model until the release of DeepSeek.
Even more surprisingly, it was produced by a Chinese company. Previously, the United States had set export controls on the most computationally powerful and advanced Nvidia Graphical Processing Units, specialized computer chips used to train the premier Large Language Models. This meant that while American AI labs such as Anthropic, OpenAI, Meta and Google had access to a vast pool of computing power, foreign labs didn’t have those same luxuries, experts said. Deepseek challenged these perceptions surrounding America’s domination over AI platforms.
Immediately, panic spread across the American tech industry, exemplified by over a trillion dollars in market capitalization being wiped out in a day. There was an almost unanimous consensus that DeepSeek was a dangerous technology, acting as an existential threat to the future of American AI. Critics highlighted the potential for Chinese propaganda powered by AI, national security concerns and the invasion of Americans’ privacy as dangerous aspects of DeepSeek.
DeepSeeks’ most significant impact transcended any technological or financial response. Instead, American competitors’ reactions proved to be the most critical shift. It was compared to the launch of Sputnik, another instance where American technological superiority was challenged. Sputnik was the first human satellite to reach orbit that the Soviet Union launched. It triggered the space race and initiated a nationwide effort to win that race. Similarly, this moment triggered a reaction to maintain that dominance. DeepSeek provided a much-needed wake-up call that we should celebrate.
The first way the U.S. responded was through a massive increase in investment. Stargate was officially announced 11 days after the release of DeepSeek. The new venture is a $500 billion investment in building infrastructure for AI over the next four years. This shift will allow AI models to become more advanced, affordable and accessible, further democratizing this critical technology. By removing the current paywalls that gatekeep the most capable models, more Americans can try these models for themselves.
On the research side of this technology, DeepSeek claims that they can train a model with only about $6 million of computing power. Astonishingly, comparative models are estimated to cost up to $100 million. By open-sourcing their research, they allowed other companies insight into their techniques. These innovations are expected to reduce AI’s energy consumption, lightening the technologies’ impact on climate change.
DeepSeek’s novel architecture and training methods allowed them to offer their best-performing model at a fraction of the cost of other companies. This meant that all users, regardless of a subscription, could use these models without limit. Whether in reaction to DeepSeek or by pure coincidence, 13 days after R1’s launch, OpenAI announced that they would offer their newest reasoning model to free tier users. Elon Musk’s AI company, xAI, released its frontier model a month later, arguably becoming the top model, according to various benchmarks.
Although many see DeepSeek as adversarial product or untrustworthy, beyond that shallow view, it is evident that the true impact of this alarming technology is how we respond to it.