US Must Embrace Open Source AI to Compete With China
▼ Summary
– The US leads in advanced AI models from companies like OpenAI and Google but is falling behind in open-weight models that can be downloaded and run locally.
– Chinese companies such as DeepSeek and Alibaba are gaining global popularity with their open models, which are more capable and developer-friendly than current US open offerings.
– Nathan Lambert founded the ATOM Project to warn that US reliance on foreign open models risks innovation and security if those models become unavailable or closed-source.
– Open models foster innovation, allow sensitive data handling on private hardware, and are essential for the US to lead in AI research and diffusion.
– Meta initially spurred the open-source AI movement with Llama in 2023, but US firms have since prioritized superhuman AI development over openness, while China has increased its open model releases.
The United States currently holds a dominant position in artificial intelligence, largely driven by sophisticated proprietary systems from firms such as OpenAI, Google DeepMind, Anthropic, and xAI. However, a critical shift is underway in the global AI landscape, with open-weight AI models becoming a central battleground for technological leadership. While American companies have excelled in creating powerful closed models, they are now being outpaced by Chinese competitors like Kimi, Z.ai, Alibaba, and DeepSeek, which are releasing highly capable open models that anyone can download, modify, and operate on local hardware.
Nathan Lambert, founder of the ATOM (American Truly Open Models) Project, emphasizes the strategic importance of this development. He states that for the U.S. to maintain its advantage across the entire AI ecosystem, it must actively support and produce top-tier open models. Currently, the most advanced American AI systems are only accessible via chatbot interfaces or APIs, keeping the core technology locked away on corporate servers. Although companies like OpenAI and Google have released some open-weight versions, these are significantly less powerful and less adaptable than the open models emerging from China.
Chinese AI firms are reaping substantial benefits from their open-source approach. By allowing external researchers and developers to access, experiment with, and improve their models, they create a collaborative innovation loop. The best external modifications and ideas can then be integrated into future model releases, accelerating overall progress. This stands in stark contrast to the current trajectory of many leading U.S. AI labs, which are increasingly focused on a closed, proprietary race toward superhuman-level artificial intelligence.
Lambert, who also serves as a researcher at the Allen Institute for AI (Ai2), established the ATOM Project to spotlight the national security and economic risks of falling behind in open-source AI. He argues that reliance on foreign open models poses a significant threat; if those models were suddenly made proprietary or discontinued, it could severely disrupt American companies and research institutions. Furthermore, open models are essential for fostering innovation among startups and academic researchers who lack the resources to build foundational models from scratch.
Businesses handling sensitive data, particularly in sectors like finance, healthcare, and defense, also have a critical need for open models. These organizations require AI systems they can run and audit on their own secure, on-premises hardware, without depending on external cloud services. Lambert asserts that open models form the bedrock of AI research, diffusion, and practical application, and the U.S. must lead in this domain rather than follow other nations.
The ATOM Project, launched on Independence Day, makes a powerful case for increased openness, documenting how Chinese open-weight models have surpassed their American counterparts in both capability and developer support in just a few years. Ironically, the modern open-source AI movement was initially ignited by the U.S. company Meta when it released its Llama model in July 2023. This model was an immediate hit, providing a vital tool for a global community of researchers and entrepreneurs and demonstrating the power of open access.
However, the strategic focus of Meta and other major U.S. players has since shifted. The intense competition to be the first to achieve artificial general intelligence has led to a retreat from openness. Meta has recently restructured its AI division, making high-profile hires and launching a new “superintelligence” lab. In a notable change of direction, CEO Mark Zuckerberg has hinted that the company’s most advanced future models may not be open-sourced.
Meanwhile, China’s tech industry has been moving decisively in the opposite direction. In January 2025, the startup DeepSeek made global headlines with its DeepSeek-R1 model. This open model was not only remarkably capable but was also trained for a fraction of the cost of leading U.S. systems. Its release signaled a new era of accessible, high-performance AI from China, prompting a wave of other Chinese firms to introduce their own powerful and innovative open-weight models, further solidifying their lead in this crucial field.
(Source: Wired)
