Meta, Microsoft, and Nvidia Unite for the Future of Open Weight AI

A Collective Call for Open Weight AI

In a collective initiative, twenty-four companies and organizations have signed an open letter addressed to American policymakers, advocating for the protection of open weight artificial intelligence (AI) models. Published today, this letter gathers signatures from tech giants such as Meta, Microsoft, Nvidia, and IBM, as well as other influential players like Dell Technologies, CrowdStrike, Palantir, and ServiceNow. Organizations like Hugging Face, Perplexity, Mistral, Andreessen Horowitz, Y Combinator, Linux Foundation, and Mozilla are also among the signatories.

What makes this coalition notable is the diversity of the signatories, which includes direct commercial competitors as well as organizations with little obvious overlap in their business models. The central argument of this letter draws an analogy to the open-source software movement of the 1980s. At that time, the question was whether source code should be accessible to all or restricted to commercial uses. Today, a similar debate is emerging regarding the weights of AI models: should they be freely accessible or confined behind commercial APIs?

The Importance of Open Weight Models

Open weight AI models are characterized by the availability of their trained parameters, allowing anyone to download, examine, modify, and run them on their own hardware. This contrasts with closed models, such as those offered by OpenAI or Anthropic, which are only accessible via APIs, with the weights remaining confined to the provider's infrastructure.

The signatories of the letter argue that open weights are essential for spreading AI capabilities beyond well-funded labs, towards practical applications in various sectors such as factories, hospitals, farms, classrooms, and local businesses.

Their argument unfolds in three main points:

• Open weights lower the entry barrier for startups and public institutions that cannot afford to develop cutting-edge models or pay high fees for using models via APIs.

• They stimulate competition at all levels, from chips to cloud infrastructure to applications, which, according to the letter, keeps costs low and prevents value concentration in the hands of a few providers.

• Finally, open weights provide enterprise customers the ability to avoid vendor lock-in, as organizations using these models can control their own data and tailor the model to their specific needs without relying on a single provider's decisions.

Security and Open Models: An Inverted Perspective

One of the most striking arguments in the letter concerns security. The signatories acknowledge that publishing a model's weights means losing control over their use. Modified versions can circulate without security safeguards, and there is no mechanism to recall them.

However, they argue that prohibition is not the solution. By comparing it to cybersecurity, they assert that defenders need access to AI models comparable to those of attackers to detect and simulate threats, which closed systems do not easily allow.

They also contend that closed models are not inherently more secure, as they can be hacked or misused, and their failures cannot be observed or verified by external researchers. In contrast, open models allow these researchers to examine behavior, conduct thorough testing, and identify vulnerabilities, thereby enhancing overall security.

The letter draws a parallel with the argument that "open-source is more secure than obscurity," although it does not provide specific data on vulnerability discovery or incidents to support this claim in the context of AI systems.

Distillation: A Controversial Technique Defended

The letter also addresses the issue of distillation, a controversial technique in the AI field. Distillation involves using the outputs of one model to train or improve another model. It is a common practice in machine learning research and development, used to evaluate, validate, and transfer capabilities between models of different sizes.

The signatories distinguish legitimate distillation from "illegal efforts to extract value from closed models," asserting that the former should not be restricted by the same limitations as the latter.

This seems to be a direct response to disputes that erupted following the rise of Chinese models like DeepSeek and Kimi, when several American labs suggested that rival models had been trained by distilling the outputs of their own closed systems without permission.

The letter proposes addressing the hijacking through targeted legal and commercial mechanisms, rather than through blanket restrictions on a technique essential to the field.

Implications for the Upcoming Political Debate

Although the letter does not come with any specific legislative or regulatory proposals, it serves as a strategic positioning ahead of anticipated political action on AI regulation in Washington. The signatories urge lawmakers to broaden access to computing for startups and researchers, fund shared training datasets and evaluation frameworks, and avoid "premature restrictions" on open models.

This document should be seen as an indicator of the preferences of major infrastructure and chip providers, such as Nvidia, IBM, and Dell, who have direct business interests in the prosperity of open weight ecosystems. A broader range of deployable models translates into more sales of services and computing, regardless of which lab produced the weights.

Procurement teams evaluating the use of open weight models versus closed models should keep in mind that the political environment favoring one approach over the other remains uncertain. Any restrictions on distillation or open publication could alter the economics of self-hosted AI within a single legislative cycle.