What is DeepSeek, the Chinese AI Startup Disrupting the Global Tech Industry?
DeepSeek, a little-known Chinese startup, has made waves in the global tech industry with the launch of an AI model rivaling those developed by Google and OpenAI.
According to its creator, the DeepSeek-R1 model was built using fewer and less advanced computer chips compared to the cutting-edge technology employed by US tech giants. In a recently published research paper, the team behind DeepSeek disclosed that they spent under $6 million on computing power to train the model—an amount that pales in comparison to the multibillion-dollar budgets of companies like OpenAI, Alphabet, and Meta.
Marc Andreessen, a prominent venture capitalist, described the launch of DeepSeek-R1 as “AI’s Sputnik moment.”
The unexpected emergence of a small Chinese startup capable of competing with Silicon Valley’s top players has raised questions about US dominance in AI. It has also sparked concerns about the inflated market valuations of tech giants such as Nvidia, Alphabet, and Meta. On Monday, Nvidia, which dominates the market for generative AI semiconductors, saw its stock plunge by 17%, wiping out nearly $600 billion in market capitalization.
US President Donald Trump, who recently announced a $500 billion AI initiative involving OpenAI, Oracle, and Japan’s SoftBank, called DeepSeek’s success a “wake-up call” for US industries to remain “laser-focused on competition.”
What is DeepSeek?
Founded in late 2023 and headquartered in Hangzhou, DeepSeek was established by Liang Wenfeng, a serial entrepreneur who also runs the hedge fund High-Flyer. Although largely unknown outside China, Liang has a track record of combining emerging technologies with investment strategies.
In 2013, he co-founded Hangzhou Jacobi Investment Management, an AI-driven trading firm. Speaking to Chinese media outlet Waves in 2023, Liang rejected claims that AI development is prohibitively expensive or that startups lack opportunities in the field.
“Reproducing existing models is relatively inexpensive; minimal training or fine-tuning, based on publicly available research and open-source code, is often sufficient. However, original research demands significant computational resources, extensive experiments, and top-tier talent,” Liang said. He added that his interest in AI stems primarily from “curiosity.”
He further hypothesized that human intelligence might be rooted in language, suggesting that human thought processes could be a linguistic phenomenon. According to Liang, this implies that human-like artificial general intelligence (AGI) could potentially emerge from large language models.
DeepSeek has yet to respond to requests for comment.
On Monday, Gregory Zuckerman, a journalist for The Wall Street Journal, noted that Liang had written the preface to the Chinese edition of his book on the late American hedge fund manager Jim Simons. Zuckerman remarked, “Simons’ influence clearly runs deep.”
Why has DeepSeek disrupted the tech world?
DeepSeek’s breakthrough has called into question the strategies of both Silicon Valley and the US government.
For years, US tech firms have been perceived as having an insurmountable edge in AI, thanks to their size, ability to attract global talent, and vast investments in advanced computing infrastructure. DeepSeek’s success challenges the assumption that billions of dollars are required to lead in AI development.
“OpenAI was founded 10 years ago with 4,500 employees and $6.6 billion in capital. DeepSeek, in contrast, launched less than two years ago with just 200 employees and under $10 million in funding,” noted Adam Kobeissi, founder of The Kobeissi Letter.
DeepSeek’s engineers revealed they used approximately 2,000 Nvidia H800 chips—less advanced than top-tier chips—to train their model. By leveraging multiple specialized models, they were able to optimize the performance of these slower chips.
This development has raised doubts about the US government’s strategy to contain China’s AI progress by restricting access to high-end semiconductors. DeepSeek’s research suggests that cutting-edge chips may not be essential for developing high-performing AI models, or that Chinese firms can still access necessary components despite restrictions.
Liang disclosed in his Waves interview that DeepSeek stockpiled 10,000 Nvidia A100 GPUs before export bans were enforced in 2023.
Implications for the global AI race
While some analysts argue that DeepSeek-R1 is on par with ChatGPT—or even superior for certain tasks—the AI landscape is evolving rapidly. OpenAI’s CEO, Sam Altman, announced plans to release its next reasoning AI model, o3 mini, soon. Altman acknowledged DeepSeek-R1’s achievements but emphasized his company’s continued focus on scaling computational power.
Meanwhile, experts have pointed to the broader lessons from DeepSeek’s success. Tanishq Abraham, former research director at Stability AI, highlighted China’s ability to innovate despite restrictions on GPU access. Tara Javidi, co-director at UC San Diego’s Center for Machine Intelligence, expressed optimism about the global progress in AI research, emphasizing the need for increased investment in the US.
Tim Miller, an AI professor at the University of Queensland, suggested that DeepSeek’s success illustrates how resource constraints can foster innovation. “By lacking access to cutting-edge computational power, they were forced to find smarter solutions. Necessity truly is the mother of invention.”
While DeepSeek’s rise marks a significant milestone, it does not necessarily mean China has won the AI race. However, it underscores the importance of innovation, adaptability, and a reevaluation of the strategies employed by both governments and companies in the AI sector.
