DeepSeek Disrupts AI Leaders’ Spending Strategies

Chinese AI firm DeepSeek has positioned itself as a formidable contender against U.S. AI giants, showcasing breakthrough models that promise performance on par with industry leaders like GPT-4 but at a significantly lower cost. Since the launch of its mobile app in early January, DeepSeek has surged to the top of App Store charts in key markets, including the U.S., UK, and China. However, skepticism remains about the validity of its claims.

Founded in 2023 by Liang Wenfeng, former head of AI-driven quant hedge fund High-Flyer, DeepSeek has adopted an open-source approach, integrating a reasoning feature that explains its thought process before delivering responses. This transparency has drawn attention, but Wall Street’s reactions are divided.

Wall Street’s Mixed Reactions

Jefferies highlights that DeepSeek’s cost-efficient methods challenge the recent surge in capital expenditures (capex) by tech giants like Meta and Microsoft, each committing over $60 billion this year. The firm suggests that if smaller models like DeepSeek’s can deliver comparable performance, it could pressure AI players to justify their escalating spending, potentially slowing data center revenue and profit growth.

Citi, however, questions whether DeepSeek’s achievements were possible without advanced GPUs, noting that U.S. companies’ access to cutting-edge chips remains a competitive advantage. Meanwhile, Goldman Sachs sees broader implications, suggesting that DeepSeek’s innovations could lower barriers to entry, reshaping competition between established tech giants and startups.

Analysts Weigh In

• Jefferies: DeepSeek’s efficiency challenges the capex euphoria following recent spending announcements by Meta and Microsoft. If smaller models can perform well, it could benefit smartphone makers, though the near-term outlook for AI smartphones remains unchanged. Jefferies also notes that China’s focus on LLM efficiency, driven by chip constraints, could accelerate innovation if U.S. restrictions ease.
• Citi: While DeepSeek’s achievements are impressive, Citi doubts they were accomplished without advanced GPUs. The firm believes U.S. companies will continue to rely on advanced chips for their cost efficiency at scale.
• Bernstein: DeepSeek’s success is not a “miracle,” and the panic over its impact is overblown. The firm views innovations like DeepSeek’s as necessary for AI’s continued progress, emphasizing that efficiency gains often lead to increased demand.
• Morgan Stanley: If DeepSeek’s claims hold, generative AI could eventually run on smaller devices, boosting demand for chips and related products.
• Goldman Sachs: DeepSeek’s developments could intensify competition between well-funded tech giants and startups, with a shift from training to inference and potential global expansion for Chinese AI players.
• J.P. Morgan: DeepSeek’s cost reductions are notable, but its reliance on High-Flyer’s GPU resources raises questions. The firm’s disruptive approach challenges the notion that higher investment guarantees innovation.
• UBS: Lower AI training and inference costs could spur broader adoption, particularly among retail customers, potentially shifting demand to data centers in tier 1 cities and improving pricing power for operators.

Conclusion

DeepSeek’s rise has sparked a lively debate among analysts, with some viewing its cost-efficient models as a game-changer and others questioning the feasibility of its claims. While the long-term impact remains uncertain, DeepSeek’s innovations underscore the evolving dynamics of the AI industry, where efficiency and accessibility could redefine competition and adoption.