In a recent announcement that has sent ripples through the tech community, Chinese startup DeepSeek has unveiled its latest AI model, R1, claiming capabilities that rival those of industry giants like Google and OpenAI, Al Jazeera reports.
The bold assertion that DeepSeek developed R1 at a fraction of the cost — some $5.6 million — has raised eyebrows and ignited debates over China’s growing prowess in the AI sector and the previously unchallenged market dominance of US tech firms.
DeepSeek, founded by serial entrepreneur Liang Wenfeng in late 2023, stated in a research paper that it utilized 2,000 Nvidia H800 GPUs to create R1’s foundational model, known as V3. This claim stands in stark contrast to the cost estimates for similar models from top US companies; OpenAI CEO Sam Altman has indicated that training its GPT-4 chatbot exceeded $100 million, employing nearly 25,000 advanced H100 GPUs.
The revelation has led to widespread speculation about the implications for the AI landscape. Many industry observers are questioning whether the US can maintain its competitive edge in AI technology, particularly in light of Washington’s efforts to curtail China’s capabilities through stringent chip export bans.
Nvidia, which dominates the supply chain for GPUs powering generative AI, felt the immediate impact of DeepSeek’s announcement. Its shares plunged 17 percent, erasing nearly $593 billion from the company’s market value—a loss comparable to the gross domestic product of Sweden.
While there is a general acknowledgment of DeepSeek’s achievement with R1, skepticism persists regarding the company’s budget claims. Palmer Luckey, founder of Oculus VR, referred to DeepSeek’s stated spending as “bogus,” suggesting that the firm’s narrative is simplistic and potentially misleading. Alexandr Wang, CEO of Scale AI, echoed this skepticism, asserting that DeepSeek likely had access to 50,000 higher-capacity H100 chips, although he did not substantiate his claims with evidence.
Tech billionaire Elon Musk also weighed in, amplifying criticism of DeepSeek on social media.
Further complicating the picture, Liang Wenfeng previously revealed that DeepSeek had stockpiled 10,000 older Nvidia A100 chips before the Biden administration enacted export restrictions.
Despite R1’s capabilities, users have highlighted limitations stemming from the model’s Chinese origins, including censorship of sensitive topics dictated by Beijing, such as the Tiananmen Square massacre and Taiwan’s status.
As fears about DeepSeek’s disruptive potential in the tech sector start to dissipate, Nvidia’s shares rebounded slightly, rising nearly 9 percent. The tech-heavy Nasdaq 100 index also saw an uptick, climbing 1.59 percent after a notable decline.
