DeepSeek R1 shakes up the AI market

On January 20, 2025, Chinese lab DeepSeek released the open-source model R1. On mathematics, code, and reasoning it performed comparably to OpenAI's o1 — but was freely available and trained at far lower cost.

Cheap and open

DeepSeek claimed R1 was trained at a fraction of the usual cost. The model and weights were public, so anyone could run and adapt it. The DeepSeek app briefly became the most-downloaded free app in the US App Store.

Stock market shock

The release caused significant drops in Western AI and chip company stocks, with Nvidia as a notable example. Investors questioned whether the enormous investments in computing power were necessary if top performance could also be achieved cheaply.

Geopolitics and open source

R1 shifted the AI power balance: China proved to be close on the heels of the US, and open-source models proved competitive with closed top models. It accelerated the debate on export rules, chips, and AI sovereignty.