DeepSeek Could Shatter Overvalued US Stocks as America's AI Spending Comes Under Scrutiny

China's Hangzhou-based DeepSeek is laying waste to US AI stocks as investors question whether it is necessary to keep spending billions on new-age chips and power supplies for AI development, given that DeepSeek's R1 AI model was built for a relatively meagre cost of £4.49 million.

Most of the "magnificent seven" stocks declined markedly on 27th January, wiping out £86.7 billion from the fortunes of the world's 500 richest people. Industry experts are increasingly speculating if DeepSeek's AI model could lead to a major price correction for established AI giants, which have remained overvalued since ChatGPT's launch in 2022. These companies also drove 2024's year-long market rally, which buoyed the S&P 500 to record levels recently.

Experts also highlight the high capital spending by top AI players with little to show in revenue terms. The Chinese startup's R1 AI model claims to match OpenAI's o1 by multiple benchmarks for a fraction of the development costs, using fewer and less advanced chips. DeepSeek's free AI model beat ChatGPT to secure the top position on the Apple App Store download charts soon after its launch last week.

Earlier this month, DeepSeek also launched its AI assistant based on the DeepSeek-V3 model, facilitating seamless user interactions by leveraging 600 billion parameters. Analysts cautioned that the emergence of low-cost Chinese alternatives to popular US AI models could disrupt the investment landscape, compelling firms to reevaluate their AI strategies.

Former Hedge Fund Manager Behind DeepSeek

DeepSeek founder Liang Wenfeng graduated from Zhejiang University and co-founded the quantitative hedge fund High-Flyer in 2015. The quant trading hedge fund leverages AI in its investment strategies to forecast market trends and make profitable trades.

Before the Biden Administration limited US exports of AI chips to China, Wenfeng was buying thousands of Nvidia graphics processing units (GPUs), which his acquaintances and business partners deemed an AI side project.

"When we first met him, he was this very nerdy guy with a terrible hairstyle talking about building a 10,000-chip cluster to train his own models. We didn't take him seriously," Liang's business partner told the Financial Times. "He couldn't articulate his vision other than saying: 'I want to build this, and it will be a game changer.' We thought this was only possible from giants like ByteDance and Alibaba."

The hedge fund manager founded DeepSeek in 2023 to develop artificial general intelligence to make the company a homegrown AI leader, recruiting top talent from the mainland. The 40-year-old founder has become the leading figure in China's AI initiative. According to Pitchbook, High-Flyer was a listed investor in DeepSeek with a minority holding.

Making The Most Out Of Limited Resources

DeepSeek's achievement in the AI space could be due to the company's resource optimisation techniques. The R1 AI model rivals OpenAI and Anthropic's Claude but is much more cost-efficient and stands out for being open source.

Wenfeng believes that sharing DeepSeek's milestones through an open-source model provides an edge, and adopting a closed-source model won't stop rivals from catching up. He also described DeepSeek as a bottom-up company without a rigid hierarchy, fostering free collaboration and access to computing resources.

Although experts opine that Chinese firms generally have limited access to top-of-the-line GPUs due to US export controls, Scale AI CEO Alexandr Wang believes that Chinese labs have more AI GPUs than people think.