Nvidia Leads 25 Companies in Open Weights Letter; Anthropic Still Absent as Signatories Grow to 50

On July 24, 2026, Nvidia CEO Jensen Huang released a policy letter titled "Open Weights and American AI Leadership," initially co-signed by 25 companies, including Nvidia, Microsoft, Meta, IBM, Dell Technologies, Palantir, and Hugging Face. The letter urged Washington to avoid imposing premature restrictions on downloadable AI models. The number of signatories grew to 50 within a day, with OpenAI and Google joining, while Anthropic and Amazon remained absent throughout.

The Facts

The letter's core demand was opposition to "premature restrictions on downloadable AI models," emphasizing that open weights models can strengthen security, accelerate innovation, and support sovereignty. Signatories spanned chip manufacturers, server vendors, cloud operators, enterprise software companies, and venture capital firms, among which Meta, Mistral, and Black Forest Labs have already publicly released their weights. The Linux Foundation also participated, using the OpenMDW-1.1 license. The letter did not directly mention China or specific models, but the context involved the Trump administration's resumed discussions on restricting Chinese models. Anthropic CEO Dario Amodei issued a statement on July 27, saying the company had never advocated banning open weights models, describing models without dangerous capabilities as "public goods."

Mechanism Analysis

The rapid spread of the letter stemmed from 11 million views generated by Huang's initial post, with signatures continuing to accumulate while coverage was still being written. This reflects the accelerated pace of technology policy formation: the document itself is simple, and dissemination becomes the core of the event. Most signatories were either model developers who had already released open weights or companies reliant on downstream hardware and cloud services. Anthropic's absence was consistent with its long-standing positioning—the company advocates mandatory testing for models with sufficient capabilities rather than blanket bans, though this stance was described by White House AI adviser David Sacks as a "fear-based regulatory capture strategy." The letter also called for expanding compute access for startups and researchers, as well as promoting investment in public datasets and evaluation frameworks.

Industry Impact

In terms of the competitive landscape, the 50 signatories span a multi-layered supply chain from chips to applications, strengthening the accessibility of open weights in enterprise clusters, regional clouds, and on-premises deployments. Developers can download weights directly to run models without relying on a few hyperscale API endpoints, lowering the threshold for procuring specific hardware. Enterprise users gain more sovereignty options but must handle data provenance auditing and agent behavior liability themselves. Anthropic's absence has put its safety-first image at odds with internal sentiment in Silicon Valley, with some reports suggesting this could affect the depth of its collaboration with signatories. Server and cloud vendors may see hardware sales growth driven by the distributed deployment demand created by open weights, while laboratories focused on closed frontier models must contend with the policy debate shifting direction.

Comparisons and Precedents

The letter noted that Huang pointed out during CES 2026 that a quarter of currently generated tokens come from open models, a proportion that coexists alongside closed models. The Linux Foundation's participation, citing the release of Nemotron 3 Ultra as a case, shows that open licenses are already being applied to real-world models. Google and OpenAI's shift from initially not signing to joining indicates a change in policy winds. Anthropic's response emphasized that its position overlaps with the letter on chip controls and distillation prevention, but the public disagreement is still interpreted externally as a conflict of commercial interests.

Strategic Assessment

Based on the current signing dynamics, the most likely scenario is that more infrastructure and application-layer companies will join the open weights camp to match the distributed deployment trend. Developers should prioritize evaluating whether model weights are publicly available, whether licenses permit local operation, and whether downstream governance tools are in place; enterprises must weigh open control against audit responsibility and establish internal compliance processes. Validation signals include whether subsequent policy documents adopt the letter's recommendations on compute sharing and dataset investment, and whether Anthropic adjusts its public stance.