DeepSeek headlines reads ‘negative’ for hyperscalers, says Raymond James

Media headlines are shining light on DeepSeek’s R1, a China-made large language model, with questions around data censorship, sovereignty, performance and competitive launches, though the headlines likely read initially negative for hyperscalers Amazon (AMZN), Alphabet (GOOG) (GOOGL), and Meta (META) provided potential competitive and return on invested capital pressures, Raymond James tells investors in a research note. The firm notes that like OpenAI, DeepSeek endorses increased test-time compute for advanced reasoning, which could shift scaling laws towards inference compute, and thinks the AI race is still in the early innings with hyperscalers well-positioned with respect to compute/power, R&D, distribution, development environments, and commercialization advantages.