China’s DeepSeek: Bold Claims Shaking the AI Industry

Questions Arise Over DeepSeek’s Bold Claims

In the fast-evolving world of AI, China’s DeepSeek has emerged as a disruptive force, directly challenging industry leaders like Google and OpenAI. The Hangzhou-based startup unveiled R1, sparking intense debates in Silicon Valley and beyond.

Many see the United States’ dominance in AI as deeply tied to market valuation and cutting-edge innovation. However, some sceptics question how DeepSeek managed to operate on a limited budget, securing funding while leveraging advanced chips. Critics argue that these assumptions require deeper scrutiny, with experts like Pedro Domingos, a professor at the University of Washington, expressing doubts.

Training such an AI model involves significant costs, and many suspect that DeepSeek’s fine-tuning and post-processing were built on more expensive resources than disclosed.

DeepSeek’s research and development team claims to have used Nvidia H800 GPUs, designed under US export controls, spending just $5.6m to train its foundational model, V3. In contrast, OpenAI’s CEO, Sam Altman, stated that GPT-4 required over $100m and nearly 25,000 H100 chips. This groundbreaking announcement sent shockwaves through the industry, with entrepreneur Liang Wenfeng placing DeepSeek at the forefront of AI innovation.

While major companies pour billions into data centres and high-end infrastructure, DeepSeek’s strategy raises concerns over Washington’s constraints on exports and their effectiveness in limiting China’s AI sector. The impact was immediately felt in California, where Nvidia, holding a near monopoly in generative AI chips, saw its stock drop 17 percent on Monday, erasing $593bn in market value—a figure comparable to Sweden’s GDP.

While there is broad consensus that DeepSeek has made a major achievement with the release of R1, industry observers remain skeptical, closely analyzing the claims made by the Chinese AI startup.