The American tech industry is growing increasingly anxious over China’s strides in open-source artificial intelligence (AI). According to an August 7 report by The Washington Post, Chinese companies dominate the field, with five of the top 15 AI models being open-source and developed by Chinese firms, while the U.S. has made slow progress in this area. In response, U.S. tech leaders this week launched the American Truly Open Models (ATOM) initiative, aiming to pool resources to develop powerful open-source AI models and reclaim the competitive edge that China has taken.
Analysts, however, caution that while the initiative is ambitious, it must overcome challenges such as coordination issues and prohibitively high costs.
In Silicon Valley and Washington, politicians often reiterate that the U.S. must defeat China in the AI race to safeguard its economic and national security. However, in the realm of open-source AI models, U.S. companies have already fallen behind their Chinese counterparts.
Historically, technological revolutions like the internet and cloud computing were accelerated by open-source software. Open-source technology allows anyone to use and modify it, enabling programmers, researchers, and entrepreneurs to test and develop powerful new ideas.
But in the current AI boom, the U.S. has lagged in open-source software development.
According to Artificial Analysis, a leading global AI benchmarking firm, only five of the top 15 AI models are open-source, and all of these are developed by Chinese companies. In July alone, Chinese AI labs released four leading open-source models, while no major models were launched by U.S. developers.
Now, some U.S. executives, investors, and scholars are backing the ATOM initiative, which seeks to enhance the competitiveness of American open-source AI.
The American Truly Open Models project aims to create a U.S.-based AI lab focused on developing software that is freely accessible and modifiable by developers. To achieve this, the plan requires substantial computational resources, including up to 10,000 advanced GPUs for enterprise-level AI development.
The initiative was officially launched on August 4, with support from more than a dozen industry leaders. These include prominent tech investor Bill Gurley, Hugging Face CEO Clement Delangue, Stanford University professor and AI investor Chris Manning, Nvidia’s Director of Applied Research Oleksii Kuchaiev, OpenAI’s Chief Strategy Officer Jason Kwon, and SemiAnalysis CEO and founder Dylan Patel.
Reports indicate that OpenAI is set to release its first open-source AI model in years, following multiple delays.
The slow pace of progress in the U.S. open-source AI sector further underscores the need for the ATOM initiative. Since Meta’s launch of the Llama 4 model in April, no significant new open-source AI products have emerged from the U.S., and even Llama 4 has disappointed many experts.
Meta CEO Mark Zuckerberg recently announced the company’s new “superintelligence” project, signaling that strategic decisions about which models to open-source would be made cautiously. This has led some AI insiders to question whether Meta will continue prioritizing open-source models.
Despite frequent references to China in discussions about AI policy, the U.S. tech giants and politicians have made limited investments in building a strong domestic open-source AI infrastructure.
“This is largely a coordination issue,” said Nathan Lambert, senior scientist at the Allen Institute for AI and one of the ATOM initiative’s founders. “There are many people in the U.S. working on AI, but they have not scaled their efforts.”
Lambert emphasized that the goal of the ATOM initiative is to develop open-source AI products far more powerful than anything currently available in the U.S. He noted that while American AI teams are willing to take risks for scientific progress, a lack of funding remains a significant obstacle.
The high cost of top-tier AI systems is another challenge. Lambert estimates that acquiring 10,000 advanced GPUs would require an investment of at least \$100 million.
“If the U.S. wants to remain competitive, this funding must be secured,” he said. “Although the U.S. still has influence in open-source AI, tech leaders and policymakers must ensure we remain at the cutting edge and accept the costs involved.”
Lambert acknowledged, however, that the support for open-source research institutes remains limited.
“If a model like DeepSeek-V3 can be trained for \$5 million, I’m sure many wealthy individuals would step forward to fund it,” he said, urging U.S. tech companies, executives, government agencies, and philanthropists to back the ATOM initiative.
Irene Solaiman, Chief Policy Officer at Hugging Face, pointed out that the ATOM initiative could not only drive scientific research openness but also help resource-limited global AI startups and projects.
“The ‘DeepSeek moment’ earlier this year inspired many small developers who now want to compete globally,” she said.
Lambert warned that falling behind China in open-source AI could jeopardize the U.S.’s AI development and influence.
He noted that data from Hugging Face suggests that despite concerns over Alibaba’s Qwen model’s training data and development, AI developers are increasingly favoring it due to its power as the most robust free model available.
Beyond the tech industry, the U.S. federal government is also advancing AI-related initiatives. On July 23, the White House released an AI action plan outlining dozens of federal policy actions aimed at maintaining an edge over China. Some analysts believe this signals the Trump administration’s intent to reverse several Biden-era policies and relax AI development regulations.