Nvidia, Microsoft and Meta join push to prevent broad limits on open-weight AI models

Nvidia, Microsoft and Meta have joined a wider technology industry effort asking U.S. policymakers to avoid broad restrictions on open-weight artificial intelligence models, according to a report published July 24 by Fox Business.

Open-weight AI models are machine-learning systems whose model parameters are made available to developers, enabling others to inspect, modify, and run the models, typically under a set of terms. Supporters argue that this transparency can accelerate research and allow firms to tailor models to specific security, privacy, and compliance needs. Critics of “open” approaches often worry that widely accessible model weights could be used to build harmful applications or to evade safety controls.

In the letter referenced by the report, Nvidia and Microsoft were among the named companies that called for a more targeted policy approach rather than sweeping rules that could apply broadly to a category of models. The report also says the coalition included Meta and that more than 20 technology companies signed on.

While the companies’ public ask focused on the scope of restrictions, the underlying issue is the policy balance between access and safeguards. Firms supporting open-weight models tend to argue that blanket restrictions are likely to push development into less visible channels or complicate efforts to test and audit systems. Those calling for tighter controls generally argue that broad access can increase the speed at which risky use cases spread.

For major platform and cloud companies such as Microsoft, open-weight models matter because a large portion of enterprise AI deployment depends on the ability to customize and integrate models into internal workflows. If policymakers create restrictions that are hard to interpret or enforce, it can increase compliance friction for customers and for developers building on shared infrastructure.

For Nvidia, whose business is closely tied to AI compute hardware and the software stack that runs AI workloads, the policy direction affects demand patterns for training and inference. Model availability and the ability of developers to run and fine-tune open models can influence what kinds of workloads are created in data centers and at the edge.

The report did not provide details on the specific policy proposals, definitions, or enforcement mechanisms that the companies were addressing. It also did not disclose whether the companies asked for exceptions for certain categories of use, suggested particular thresholds for access, or proposed alternative guardrails, such as documentation or evaluation requirements.

What to watch next is whether the industry’s request changes the language in pending U.S. rulemaking or guidance, and whether regulators announcement a preference for narrow, risk-based requirements instead of broad category bans. Given that the letter includes heavyweight AI and cloud players, its influence will likely be measured in how future drafts define “open-weight” models and what exemptions, if any, are carved out for vetted developers and audited deployments.