The sell-off in tech stocks, especially Nvidia, from the rise of DeepSeek has sparked mixed reactions among analysts at major financial institutions.
Analysts from JPMorgan caution that the AI investment cycle may be overhyped, while Jefferies proposes two strategies: continue investing in computing power or focus on efficiency, which could reduce AI capital expenditure in 2026. In contrast, Bernstein and Citi downplay the panic surrounding DeepSeek, maintaining confidence in US companies like Nvidia and Broadcom.
Meanwhile, Raymond James suggests that DeepSeek’s innovations could reduce training costs and the need for large GPU clusters. Cantor, however, views these developments as bullish for GPU demand, expecting an increase in GPU needs and recommending that investors buy Nvidia when the price drops.
DeepSeek was founded in Hangzhou, China, when Liang Wenfeng, co-founder of High-Flyer, recruited the company’s research unit in April 2023 to focus on large language models and artificial general intelligence. The Chinese startup gained attention with its reasoning model, R1, which rivals OpenAI’s o1.
Despite US chip export restrictions, DeepSeek successfully developed the model. Additionally, R1 is open source, enabling developers to use it freely. With these factors and the fact that the API’s price of DeepSeek is 27 times cheaper than ChatGPT, the US AI seems less superior.
Nevertheless, DeepSeek does have one weakness that can deter foreign customers. The Chinese company now temporarily allows only those with China mobile phone numbers to register. This limitation was set after DeepSeek’s service was disrupted by a “large-scale malicious attack.”
