Each week we find a new topic for our readers to learn about in our AI Education column.
Ladies and gentlemen, we have had our first artificial intelligence market disruption of 2025, and this week on AI Education we’re going to talk about it. Let’s dive right in. On Jan. 20, a Chinese AI startup, DeepSeek, publicly launched a new chatbot linked to its latest artificial intelligence model, DeepSeek-R1. Within a week, it became the most downloaded free app in the Apple App Store, eclipsing Open AI’s ChatGPT.
But technology stocks around the world shook in response to R1 not because of its power or sophistication, but because of its efficiency. GPU maker Nvidia lost $600 billion in market value on Monday, Jan. 27 alone. The U.S. Nasdaq composite had a 3% sell-off, taking the Dow Jones Industrial Average and the S&P 500 down with it.
R1 is in and of itself not so revolutionary. OpenAI, perhaps the most progressive of the big U.S. artificial intelligence developers, has a comparably performing model in o1, which was released last year, and is at work on even more sophisticated reasoning models as R1 begins to compete. Somehow, this model that compares well to the best of what OpenAI had to offer, was developed and trained far more efficiently than any sophisticated artificial intelligence to date. DeepSeek claims R1 took two months and less than $6 million to build and train.
What Is DeepSeek
Based in Hangzhou, China, DeepSeek is an AI firm founded in 2023 by Liang Wenfeng, who co-founded the Chinese hedge fund High-Flyer. Like the U.S.-based OpenAI, DeepSeek is an open-source AI firm making its algorithms, models and training details available to the public—but unlike OpenAI, DeepSeek’s technologists are all of one nationality: Chinese. The company seemingly flew under the radar until the launch of R1 in December 2024, and didn’t really catch the public’s eye until it rolled out its R1 chatbot this month.
According to the published narrative, Liang employed artificial intelligence algorithms and trading at High-Flyer. During this time, he accumulated a store of GPUs before U.S.-led sanctions throttled the flow of AI-related chips to China. DeepSeek was first established in 2023 as an AI lab associated with High-Flyer to build out non-financial AI tools. In November 2023, it launched its first large language models.
Why Did R1 Shock the Markets
AI is widely viewed as a resource-expensive technology. It requires a lot of processing power, which means high concentrations of powerful computer chips, requiring a lot of energy for both power and cooling, and time to train and optimize models. Firms like OpenAI and Meta spend tens of millions of dollars to train models, using tens of thousands of cutting-edge GPUs. In the U.S., huge initiatives, like the recently announced Stargate, are built to pour hundreds of billions of dollars into data centers.
DeepSeek reports that R1 was trained using about 2,000 GPUs that were not representative of the latest and greatest technology. If the company is to be believed, the artificial intelligence future may be built at significantly less financial and energy expense. When DeepSeek launched it A.I. Assistant chatbot, which uses the R1 model, it became apparent to users that the quality of the AI was on par with what OpenAI was offering in its free-to-use release.
The implication to investors was that plans to build massive amounts of new energy production to power a massive number of new data centers may be premature—there may be a far cheaper and easier way to build out AI. There wouldn’t be a pressing need for next-generation AI chips. Less space, less cooling, less energy. Investors began to reconsider the value of their AI-related investments. Nvidia, long at the top of the heap of companies benefitting from the AI boom, was also the first victim of the brief DeepSeek panic of 2025.
The New AI Arms Race
DeepSeek’s R1 and A.I. Assistant chatbot are also a direct challenge to the United States and its allies. Some are even calling this a “Sputnik moment,” where we suddenly realize that we’re losing a technological competition to a perceived adversary. The irony is that U.S. policy may have created the necessity for DeepSeek’s found efficiencies—by restricting the flow of AI-related chips to China, DeepSeek’s engineers had to find ways to do more with less, thus finding a path towards the same kind of AI created by U.S. companies but with less sophisticated and less expensive technological underpinnings.
If a Chinese company can do the same work for cheaper, then the Chinese company should win more business than its Western competitors. If its consumer-facing technology offers a similar experience at less cost, then it will attract more users—and collect more data—than its counterparts in the West, and it will be more effective at building and training subsequent generations of artificial intelligence.
In other words, yes, this is like a Sputnik moment. While the West and its allies have long assumed dominance in producing semiconductors and their associated technologies, DeepSeek’s advancements, if true, may at least temporarily decouple progress in artificial intelligence from the power and quality of its underlying hardware. The U.S. is now somewhat behind China in developing and proliferating AI, and if it does not close the gap quickly, chances increase over time that it will.
Is It an Over-Reaction?
Markets always over-react to the news, in our opinion, however, there is a good chance that DeepSeek has built a better AI mousetrap and that it’s time to temper some of our more optimistic assumptions about the impact of artificial intelligence on economic growth. Nvidia is going to be fine—in fact, most of those companies caught in the Nasdaq’s miserable Monday decline are going to be just fine, and the negative impact of DeepSeek on financial markets will probably seem like a blip by year’s end.
But the AI landscape just entered a new phase of rapid change. Even if the news around R1’s capabilities or development costs turns out to be exaggerated or untrue, U.S. policymakers are now feeling the heat of Chinese competition. Just as Sputnik accelerated the missile race, NASA and eventually saw the footprints of U.S. astronauts on the lunar surface, DeepSeek may be the wakeup call that propels an AI race—if not an arms race.
