Nvidia, Microsoft, Meta and more than 20 other technology companies and organizations are pressing the U.S. government to avoid sweeping restrictions on open-weight artificial intelligence models, arguing that broad limits could weaken innovation, cybersecurity and American competitiveness.
In a joint letter released Friday, the coalition called for targeted responses to intellectual property violations rather than rules that would broadly constrain open models or the techniques used to develop them. The appeal comes as the Trump administration considers how to address security and intellectual property concerns involving foreign AI developers, including China’s Moonshot AI and its open-weight Kimi K3 model.
Open-weight models make their trained parameters available for download, allowing developers and organizations to run, adapt and fine-tune them on their own infrastructure. That distinguishes them from closed services, whose model weights and underlying systems remain controlled by their developers. Open weights do not necessarily mean that every component of a system, including its training data and source code, is publicly available.
The companies presented access to such models as an important part of a broader U.S. technology ecosystem, rather than merely an alternative method for distributing AI software. They argued that downloadable models can spread AI capabilities across industries, give organizations more control over deployment and support research into security and safety.
“Our AI leadership will be judged not by one frontier AI model,” the signatories wrote, but by the strength of an open ecosystem reaching across the economy.
The letter was signed by Nvidia, Microsoft, Meta, Dell Technologies, IBM, Palantir, Hugging Face, Mozilla, Mistral, Andreessen Horowitz and Y Combinator, among other companies and groups. OpenAI, Anthropic, Google and Elon Musk’s xAI were not among the signatories.
Nvidia chief executive Jensen Huang promoted the letter in his first post on X, saying open models can support safety, cybersecurity, innovation and technological sovereignty. Nvidia is a central supplier of the computing hardware used to train and operate AI systems, while several other signatories develop models, cloud services, enterprise software or tools for distributing and modifying open-weight systems.
The policy debate has intensified around model distillation, a common machine-learning technique in which a smaller model is trained using outputs from a more capable system. White House Office of Science and Technology Policy Director Michael Kratsios accused Moonshot AI of using distillation to develop Kimi K3 from Anthropic’s Fable 5 model. Treasury Secretary Scott Bessent also raised the possibility of sanctions against companies found to have used distillation improperly to take protected intellectual property.
The coalition did not argue that improper extraction from proprietary systems should go unpunished. Instead, it acknowledged that unlawfully obtaining value from closed models presents a legitimate concern and said authorities should rely on focused legal and commercial remedies. Its position draws a line between alleged misconduct involving a particular company or model and the broader availability of open weights and model-development methods.
That distinction is likely to remain central to the U.S. policy discussion. Open-weight systems can be inspected, customized and operated without sending data to an outside provider, features valued by researchers, businesses and governments seeking greater control. At the same time, their portability can make it harder for a developer to monitor downstream modifications or prevent deployment in jurisdictions where U.S. companies face restrictions.
The letter adds a coordinated industry voice to a governance dispute spanning national security, competition and intellectual property law. The administration now faces pressure to address suspected misuse by individual developers without imposing controls that would also affect American companies and the wider open-model community.
Sources: Tech Policy Press