$100M effort seeks to build America’s answer to China’s open-source AI dominance

American tech leaders unveiled an ambitious plan to build artificial intelligence (AI) systems that can match China’s growing control over open-source artificial intelligence, requiring access to 10,000 cutting-edge computer chips worth at least $100 million.

The ATOM Project seeks to create the first major U.S. lab dedicated to developing freely available AI models. China currently dominates this space, with five of the top 15 open-source AI systems coming from Chinese companies like Alibaba and DeepSeek.

“There are groups of people in the country that are doing it, but they haven’t been able to scale up,” said Nathan Lambert, the project’s founder and a senior research scientist at the Allen Institute for AI.

The initiative has attracted support from major tech figures. Bill Gurley, a veteran investor, signed on alongside OpenAI’s chief strategy officer Jason Kwon and Nvidia’s applied research director Oleksii Kuchaiev.

Chinese companies released four leading open-source AI models in July alone. American developers released none during the same period, according to benchmarking firm Artificial Analysis.

This gap threatens U.S. influence over global AI development. Open-source software allows programmers worldwide to freely use and modify powerful AI tools, often leading to rapid innovation.

Lambert estimates the project needs $100 million to secure 10,000 state-of-the-art graphics processing units. These specialized chips power corporate AI development and cost millions of dollars per cluster.

The Trump administration’s 2025 AI Action Plan supports open-source development as a national priority. Federal funding could reach hundreds of millions of dollars, sources indicate.

China’s advantage stems from coordinated government backing. NVIDIA CEO Jensen Huang noted that “50% of the world’s AI researchers are Chinese. We will have to compete.”

The project faces timing pressure. Meta’s last major open-source release came in April with its Llama 4 model, which disappointed some experts. Recent company statements suggest Meta may limit future open-source releases.

“The ‘DeepSeek’ moment in January in many ways sparked ambitions among smaller developers to be able to compete at a global level,” said Irene Solaiman, chief policy officer at Hugging Face.

ATOM aims to deliver frontier open models within six to twelve months. Success depends on convincing tech companies, philanthropists, and government agencies to fund multiple competing labs rather than a single massive effort.