How many startups can say their ranking on the Apple app store sparked a global sell-off of tech stocks wiping out a trillion dollars of value?
Meet Chinese startup DeepSeek, a Hangzhou-based startup founded by hedge fund manager Liang Wenfeng that has demonstrated AI reasoning models that appear to be on par with OpenAI and Anthropic, at significantly lower cost. That feat catapulted the Chinese company’s chatbot ahead of U.S. rival ChatGPT in downloads from Apple’s App Store on January 27, causing shock waves in Silicon Valley and sparking a global sell-off of tech stocks, with Nasdaq, Dow Jones Industrial Average, and S&P500 futures all falling. Nvidia stock plunged 17% that day, wiping out nearly $600 billion in value — a record loss for a U.S. company.
Constrained by lack of access to highest-end AI chips from Nvidia by Washington’s export curbs, DeepSeek has taken a different approach than its U.S. competitors. It said training one of its latest models cost $5.6 million, compared with the $100 million to $1 billion range cited last year by Dario Amodei, chief executive of AI company Anthropic. According to the technical paper it published on January 27, DeepSeek said it used a cluster of just under 2,050 graphics processing units (GPUs) from Nvidia for training — much less than the tens of thousands of chips U.S. firms are using to train similarly-sized models.
“The real lesson of DeepSeek is that we can still make vast improvements in efficiency,” Gary Marcus, a cognitive scientist and leading voice in AI, wrote in his newsletter. “When all is said and done, with the right software, we might be able to build intelligence with hundreds or even dozens of GPUs.”
While industry experts were impressed by DeepSeek’s results, Suranga Chandratillake, a general partner at venture firm Balderton Capital, cautioned against overreacting to this latest twist in the global AI race. “Amid the non-stop pontification on our imminent DeepSeek-dominated AI futures and the death of OpenAI and Nvidia (led mainly by those who, a week ago, were non-stop pontificating on our imminent Open-AI and Nvidia-dominated future and the irrelevance of all technology outside the U.S.), I think really the only thing the past week shows us is: we continue to be very early on in the foothills of machine intelligence,” he said in a LinkedIn post. “To over-index on the success of any particular method, technology or company at this stage would be foolish.”
Still, less than two weeks away from the February 10 Global AI Action Summit in Paris DeepSeek’s approach to training is being seen as a sign that U.S. AI dominance is no longer a given.
As foundational model costs plummet, new opportunities will open for new players, based anywhere.“The real opportunity, particularly in Europe, will lie in vertical AI: building applications that solve specific, high-value problems in underserved sectors,” Explain CEO’s Guillaume Liegey wrote in a LinkedIn post. “This shift changes the game for startups and investors alike. With lower LLM costs, the unit economics for vertical AI applications will improve drastically, enabling cheaper experimentation, faster scaling and better margins. DeepSeek just proved that efficient AI is the future. The question now is: who will own the verticals?”
” Very valid point!” responded Philippe Huberdeau, Secretary General of the AI Action Summit. “There is clearly lot of space between Stargate [a U.S. high-profile artificial intelligence infrastructure project, being backed by OpenAI, SoftBank and Oracle, which aims to spend $500 billion on new compute infrastructure] and DeepSeek for a third European way to leverage this industrial revolution in a responsible and sustainable manner.”
The danger is that even if Europe, and the rest of the world, opt for leveraging AI in a “responsible and sustainable manner” the U.S. government and U.S. companies will see DeepSeek’s breakthrough as a threat that needs to be countered in a no-holds barred race.
In an editorial entitled The DeepSeek AI Freakout, the Wall Street Journal wrote that “Deepseek is vindicating President Trump’s decision to rescind a Biden executive order that gave government far too much control over AI.” The Journal went on to say that “DeepSeek should also cause Republicans in Washington to rethink their antitrust obsessions with Big Tech. Bureaucrats aren’t capable of overseeing thousands of AI models, and more regulation would slow innovation and make it harder for U.S. companies to compete with China.”
But, as some point out, the significance of DeepSeek is not really about U.S. versus China. It’s about the evolution of AI. If it gets cheaper to build and use, the world could avoid a new type of digital divide, based around compute power. Just days before the so-called “DeepSeek freakout”, Clara Chappaz, France’s Minister Delegate for Artificial Intelligence and Digital Technology, cautioned during a panel in Davos moderated by The Innovator’s Editor-in-Chief, “We are living in a world where the value of AI is very concentrated in the hands of a few players that have the data, the infrastructure, the energy and the talent,” she said.
DeepSeek’s breakthrough looks poised to help level the playing field, say pundits, and, if AI gets cheaper, business and consumers will be the winners.
Chief information officers at U.S. companies told The Wall Street Journal that they are hoping that the techniques the Chinese company used to build the model can be replicated by other vendors, ultimately driving costs down for corporate tech leaders. Some corporates are already testing the effectiveness of the DeepSeek model for business use cases, both in-house and through vendors that make them available through their platforms. Smaller versions of new DeepSeek models are now available on Amazon Bedrock, the company’s AI development platform, for enterprises to test, run and use for business cases, according to the Journal’s article. Reuters reported that Microsoft this week made DeepSeek’s R1 artificial intelligence model available on its Azure cloud computing platform and GitHub tool for developers.
“DeepSeek have just opened the playing field, taking some of the power away from Big Tech and making the best performing foundational models more accessible to businesses of all shapes and sizes,” AI leader Professor Andy Pardoe said in a LinkedIn post.
It is the beginning of a new chapter in AI that began, but is unlikely to end, with DeepSeek.The next trillion dollar company is the one that can lessen dependency on Big U.S. Tech companies ” and add a new stack that protects IP and data confidentiality and privacy that’s more agile,” Adecco Group’s Frank Dias, said in a LinkedIn posting. The opportunity, he said, is “floating out there in the future. It’s waiting for someone to pluck it.”