US Launches 'American DeepSeek Project' to Build Fully Open AI Models Within Two Years

Nathan Lambert, a prominent AI researcher at the Allen Institute for AI (AI2), has unveiled "The American DeepSeek Project," an ambitious initiative to develop fully open-source AI models in the United States within the next two years. The project aims to establish a robust, global scientific AI ecosystem, serving as a strategic counterweight to China's rapid advancements in open-source AI and offering a viable alternative to building products atop leading closed models. This move comes as the global AI landscape intensifies with competing approaches to model development and accessibility.

China has recently demonstrated significant strides in the open-source AI domain, with companies like DeepSeek leading the charge. DeepSeek's R1 reasoning model, launched in January 2025, has shown competitive performance against industry leaders such as OpenAI's o3, notably achieving this with remarkable cost-efficiency. Other Chinese tech giants, including Huawei, Baidu, Alibaba, and Tencent, have also released powerful open-source models, emphasizing innovation despite U.S. export restrictions on advanced hardware. These models often feature permissive licenses, fostering widespread adoption and contributing to China's growing influence in the AI sector.

Lambert's vision for a "fully open model" extends beyond just open weights, encompassing access to training data, code, logs, and decision-making processes. This comprehensive transparency is intended to reclaim U.S. leadership in open AI and reduce the industry's reliance on proprietary, closed systems, which some argue pose accountability and trustworthiness concerns. The initiative is projected to require an investment of approximately $100 million to $500 million over its two-year timeline, highlighting the significant resources needed to compete at the frontier of AI development.

The emergence of "The American DeepSeek Project" underscores a pivotal moment in the global AI race, marked by differing philosophies on openness and control. While Chinese open-source models like DeepSeek R1 have gained considerable traction and even topped app store charts due to their accessibility and performance, they also face scrutiny regarding potential geopolitical influences. The U.S. initiative seeks to provide a trusted, transparent alternative, shaping the future of AI innovation and its broader societal implications.