DeepSeek, a Hangzhou-based AI lab, published the R1 reasoning model on an open-source licence, claiming performance comparable to leading closed Western models at a fraction of the training cost.
The release prompted sharp moves in US technology stocks, including Nvidia, on concerns that hyperscaler capital expenditure assumptions might need to be revisited if frontier capabilities can be approximated with materially less compute.
DeepSeek's technical reports describe an architecture combining mixture-of-experts routing with reinforcement learning from verifiable reward signals, distilled into smaller, more efficient inference models.
Chinese policymakers have publicly highlighted the launch as evidence that domestic AI research can keep pace internationally despite US export controls on advanced chips and equipment.