Seemingly overnight, Chinese AI lab DeepSeek has captured global attention. Its chatbot app surged to the top of both Apple’s App Store and Google Play, igniting a mix of curiosity and concern across the tech world. At the core of DeepSeek’s viral success are its highly efficient AI models—so efficient, in fact, that analysts are starting to wonder whether U.S. dominance in the AI space is sustainable, and whether the current demand for AI chips can keep pace.
A Hedge Fund’s Side Project Turned Global Disruptor
What many don’t realize is that DeepSeek began as an offshoot of High-Flyer Capital Management, a Chinese quantitative hedge fund that uses AI to guide its trading strategies. Co-founded by Liang Wenfeng—an AI enthusiast who began exploring financial markets while at Zhejiang University—High-Flyer launched the DeepSeek lab in 2023 to focus solely on AI R&D. The lab quickly spun off into a separate entity under the same name.
From its inception, DeepSeek built its own training infrastructure but, like other Chinese AI firms, has been constrained by U.S. hardware export controls. For instance, the company had to train one of its advanced models using Nvidia’s H800 chips, a lower-performance version of the H100 reserved for non-U.S. markets.
Internal Talent, External Impact
DeepSeek’s research team is known for its youth and academic prowess. It actively recruits PhDs from elite Chinese universities and even brings in individuals without technical backgrounds to help broaden the model’s understanding of real-world concepts, according to The New York Times.
Its first models—DeepSeek Chat, DeepSeek LLM, and DeepSeek Coder—launched in late 2023 but made limited waves. That changed with the release of DeepSeek-V2, a general-purpose model for text and image analysis, which impressed with its benchmark results and cost efficiency. In response, competitors like ByteDance and Alibaba slashed prices for their own models, some even making them free.
DeepSeek-V3, unveiled in December 2024, continued the momentum. Internal benchmarks showed it outperforming both open-source giants like Meta’s Llama and closed systems like OpenAI’s GPT-4o.
In January, DeepSeek introduced R1—a “reasoning” model that simulates human-like logical steps. Unlike most chatbots that quickly generate responses, R1 pauses to analyze and self-verify its answers, especially excelling in complex subjects like physics and advanced math. Though slightly slower, its reasoning-based approach offers greater reliability.
Regulation and Censorship
There’s a tradeoff, however. Being a Chinese company, DeepSeek’s AI must pass government benchmarks to ensure its outputs align with “core socialist values.” As a result, topics like Taiwan’s independence or the 1989 Tiananmen Square protests are strictly off-limits in the chatbot’s responses.
A Business Model Built on Disruption
Despite—or perhaps because of—its pricing strategy, DeepSeek remains enigmatic in terms of monetization. Many of its services are offered far below market rates, and some are free. It has reportedly declined VC funding, even amid massive investor interest. The company claims its cost-cutting is possible due to breakthroughs in model efficiency—though some experts question the accuracy of those claims.
Regardless, developers are embracing DeepSeek’s tools. While not open source in the traditional sense, its models are accessible under flexible licenses that allow commercial use. On Hugging Face alone, developers have built more than 500 variations of R1, which together have been downloaded 2.5 million times.
Industry Shake-Up and Global Reaction
DeepSeek’s impact has been dramatic. Its rise coincided with a steep drop in Nvidia’s stock in January and even provoked public remarks from OpenAI CEO Sam Altman. The U.S. Commerce Department responded by banning DeepSeek from all staff devices. Meanwhile, Microsoft made DeepSeek models available on its Azure AI Foundry, underscoring both the tool’s popularity and the tensions surrounding its origins.
Criticism has come fast. In March, OpenAI described DeepSeek as “state-subsidized” and “state-controlled,” urging the U.S. to consider banning its models outright. Meta CEO Mark Zuckerberg, addressing questions during an earnings call, stated that AI spending would remain a core strategic priority in light of rising competition. Nvidia CEO Jensen Huang, however, praised DeepSeek’s innovation, noting that its reasoning models drive demand for more powerful chips—good news for Nvidia.
Still, the backlash continues. South Korea and New York State have both barred the use of DeepSeek on government devices. According to The Wall Street Journal, a U.S. federal ban is likely imminent.
What’s Next?
DeepSeek’s future remains uncertain. More advanced models are almost certainly on the way, but geopolitical tensions could limit its international footprint. Whether it continues to operate as a disruptive force or becomes a cautionary tale of regulation versus innovation is still up for debate.
One thing is clear: DeepSeek isn’t just another AI lab—it’s a flashpoint in the evolving power struggle over global AI leadership.