The DeepSeek market shock refers to the sharp sell-off in AI-related stocks, most severely Nvidia, that followed the late-January 2025 release of DeepSeek-R1, a reasoning model from the Chinese lab DeepSeek that appeared to match leading Western models on several benchmarks despite claims of a far smaller training budget. On January 27, 2025, Nvidia's market capitalization fell by roughly $600 billion in a single trading day, at the time reportedly the largest one-day loss of market value for any company in history, as investors reassessed assumptions underpinning the AI infrastructure buildout.
Background
DeepSeek, founded by Liang Wenfeng and originally spun out of a Chinese quantitative hedge fund, had released earlier models through 2024 but drew comparatively little attention outside specialist circles. DeepSeek-R1, released as an open-weights model in January 2025, matched or approached the performance of OpenAI's o1 on several math and coding benchmarks while its accompanying technical report described training techniques emphasizing efficiency, including a mixture-of-experts architecture and large-scale Reinforcement learning on verifiable tasks, and claimed a training cost dramatically lower than figures commonly associated with comparable Western frontier models.
Market reaction
The scale and low reported cost of DeepSeek-R1's training run challenged a core assumption behind years of surging capital expenditure on AI infrastructure, that frontier-level performance required massive GPU clusters and correspondingly enormous spending, an assumption that had driven Nvidia's valuation to among the highest in the world. The panic was amplified by DeepSeek's app briefly topping download charts on Apple's US App Store, suggesting rapid mainstream adoption of a low-cost alternative. Nvidia, along with other companies tied to AI infrastructure spending such as power and data center firms, saw sharp declines before markets partially recovered in subsequent weeks as some analysts questioned whether DeepSeek's reported training costs told the full story, excluding prior research expenditure and underlying hardware costs.
Aftermath and debate
The episode intensified debate over the actual costs of training frontier models and whether continued scaling required ever-larger capital investment or could be substantially offset by algorithmic efficiency gains, feeding into ongoing discussion of Scaling laws. It also sharpened attention on US-China competition in AI, given that DeepSeek had reportedly achieved its results using Nvidia chips of restricted or lesser capability due to US export controls, raising questions about the effectiveness of those controls. Following the shock, DeepSeek-R1's open weights led to a wave of derivative and distilled models built by other developers, and the event became a widely cited reference point in discussions of open-weights competitiveness against closed frontier labs such as OpenAI and Anthropic.