Last week, President Trump unveiled the U.S. AI Action Plan, highlighting “encourage open-source and open-weight AI” as a key priority, reflecting an urgent national focus and strategy in the AI race against China. Open source is essential, as shown in China’s own Action Plan, emphasizing its importance in the global landscape.
Earlier this year, China released DeepSeek-R1, a potent open-source large language model, without fanfare but with open weights and open science, allowing adaptation and innovation. Researchers quickly embraced it, making it popular on Hugging Face, and prompting widespread use across major firms, including those in the U.S., marking a shift towards Chinese-developed AI.
The U.S. stock market experienced a drop following DeepSeek’s release, with many Chinese research groups advancing rapidly in open-source AI, sharing their methodologies and models. In contrast, U.S. firms have tightened their models, with leading ones like GPT-4 only available through restricted interfaces, limiting access to their inner workings.
Once the leader in open-source AI, the U.S. is now losing its position. American developers increasingly rely on Chinese open models due to restricted access to domestic models. This shift could hamper the U.S.’s AI progression, as foundational AI work relies on open-source foundations for innovation and speed.
Open-source AI is not only vital for innovation but also for security, science, and democratic processes. Transparent models allow broader adaptations and reduce reliance on proprietary systems. The U.S. needs to bolster its open-source initiatives, as seen with Meta’s Llama and the Allen Institute for AI, to rejuvenate its standing in the global AI realm.
Clément Delangue advocates for a shift towards open-source AI, aligning with democratic values and rekindling America’s leadership in the field through open collaboration and competition.