🌅 Today’s stat: $5.6 million

SOURCE

DeepSeek

WHAT TO KNOW

• DeepSeek’s V3 is an open-source AI model the company claims is on-par with popular models like OpenAI’s ChatGPT and Anthropic’s Claude. The cost to train V3, which doesn’t include the cost to train earlier models, is a fraction of what it cost to train similar ChatGPT and Claude models, and a small percentage of the more than $1 billion researchers have said it could cost to train AI models in the future.

WHY IT MATTERS

• DeepSeek’s model calls into question the dominance of large U.S. AI companies and the hundreds of billions of dollars in planned investments they have lined up. The low cost could weaken the moat large amounts of funding provides for bigger companies, while the open-source nature could upend their pricing (OpenAI, for instance, charges $200/month for the most expensive version of ChatGPT).

CONNECT THE DOTS

• DeepSeek’s latest release—of its R1 model—came the same day President Trump signed an executive order aimed at eliminating loopholes in existing export controls, which currently include a ban on selling AI chips to Chinese companies. Experts say the timing of R1’s release is as much about the politics as it is the technology, since an important goal of Chinese foreign policy is to show U.S. export controls are ineffectual or counterproductive, likening the release to U.S.-sanctioned Chinese company Huawei’s release of a new phone during diplomatic discussions over U.S. export controls in 2023.