The release of DeepSeek’s R1 model on the eve of President Donald Trump’s inauguration wiped a trillion dollars from the stock market. Panicked investors sold Apple, Broadcom, Meta and Nvidia stock, but there has since been some recovery.

DeepSeek’s founder, Liang Wenfeng, is a Chengdu-based quant hedge-fund manager and AI engineer who bought GPUs because he was curious about the boundaries of AI. DeepSeek is said to rival models such as OpenAI’s ChatGPT and Microsoft Copilot, and it used a significantly smaller budget. It only took $5.6 million, and two months, to release a state-of-the-art model compared to OpenAI’s o1, which cost about $100 million to train, and Llama 3, which cost more than $720 million.