Nvidia and Microsoft warn against open source AI ban
Nvidia, Microsoft, IBM and Meta backed a letter urging policymakers to avoid broad restrictions on open-source AI models.
By Dominic Okoye · Staff Writer
· 3 min read
Nvidia, Microsoft, IBM and Meta are pushing back against a possible open source AI ban, backing an open letter that urges policymakers not to impose broad restrictions on open-source artificial intelligence models. The letter, published on Nvidia’s website and signed by more than a dozen other technology companies and several industry nonprofits including The Linux Foundation, frames open models as part of U.S. AI competitiveness rather than a category to be blocked.
The intervention follows reporting from Axios that some Trump administration officials wanted to limit the use of open-source AI models. Axios reported that the U.S. Commerce Department last year considered blocking access to software from Chinese companies, which have produced many of the world’s widely used open-source AI models.
The signatories argue that U.S. AI strength should be measured by the spread of open AI systems across industries, not only by the performance of a single frontier model. That is a direct policy claim from companies with commercial interests across chips, cloud infrastructure, enterprise software and model distribution. The letter does not say what specific legislation or executive action the companies expect, and it does not name a particular proposed ban.
Why are tech companies opposing an open source AI ban?
The companies say open-source models reduce dependence on proprietary large language models, lower the cost of building AI applications and let users adapt models for their own workloads. They also point to private deployment as a reason enterprises may prefer open models, since companies can run them on infrastructure they control rather than relying only on external model providers.
That argument cuts across several markets. For Nvidia, broader model availability can support demand for AI chips and systems. For cloud providers and software vendors, open models can increase competition across infrastructure, developer tools, applications and services. The letter says open weights can create more rivalry among model makers and companies selling the surrounding AI stack, which the signatories argue can lower costs and spread AI benefits more widely.
Open weights generally refers to making a model’s learned parameters available for others to use, adapt or deploy. In practice, that can let developers customize models without rebuilding them from scratch, though the letter does not disclose any benchmark data or cost comparisons for the economic claims it makes.
What risks does the letter address?
The letter acknowledges that open-source AI models can be harder to monitor once released. That creates problems for enforcement and cybersecurity, including the possibility that attackers could use advanced open models to accelerate cyberattacks.
The signatories counter that openness can also help security work. They argue that when model materials are public, a wider set of researchers can inspect them for vulnerabilities, which can reduce the number of unpatched flaws available to attackers. That is a familiar open-source software argument applied to AI models, although the letter does not provide a specific security framework for handling misuse.
The letter also addresses model distillation, a technique in which researchers use outputs from a more capable model to train a smaller one. A White House official recently said Kimi K3, an open-source model with advanced coding abilities, was distilled from Anthropic’s Fable 5 model.
The signatories say unlawful extraction from closed models is a legitimate concern, but argue that it should be handled through narrower legal and commercial tools rather than broad limits on distillation techniques. They also recommend that regulators improve researcher access to compute and free training resources such as datasets.
For AI companies, the policy fight is not only about openness. It is about who controls distribution, infrastructure spend and the pace at which competitors can build on existing models. The letter puts major U.S. technology vendors on record against a sweeping restriction, while leaving room for targeted rules around misuse and intellectual property.
This story draws on original reporting from SiliconANGLE.
