The Coalition Map: Who Signed Jensen Huang's Open-Weights Letter and Who Didn't

July 25, 2026

Jensen Huang waited years to post on X. When he finally did, on Thursday morning, he didn't share a keynote clip or a chip announcement. He shared a policy letter, "Open Weights and American AI Leadership," signed by 25 companies and organizations, arguing that America's AI future depends on models anyone can download, inspect, and run.

The post has pulled nearly 50 million views in under 48 hours. But the letter itself is only half the story. The more revealing document is the signature block at the bottom, and the names that aren't in it.

First, the timing

This letter didn't appear in a vacuum. It landed four days after reports that Washington is reviving a push to restrict Chinese open-weight models: the DeepSeeks, Qwens, and Kimis that now dominate open-model leaderboards. The letter never names China, but its plea to avoid "premature restrictions on open models that stifle competition or drive innovation overseas" is aimed squarely at that debate.

There's a second, quieter target. The letter goes out of its way to defend distillation, the practice of training one model on another model's outputs, as "a widely used technique for model improvement" that shouldn't be conflated with misappropriation. That's a direct echo of the OpenAI and DeepSeek dispute from last year, when OpenAI accused DeepSeek of distilling its models. Twenty-five companies just went on record saying distillation is legitimate engineering, not theft.

The map: who signed

Read the signatory list as a map of who profits from an open AI ecosystem, and it snaps into focus.

The hardware bloc: Nvidia, Dell. The simplest reads on the list. Closed frontier models concentrate compute demand inside a handful of labs and their cloud partners. Open weights put GPU demand everywhere: every enterprise running Llama on-prem, every startup fine-tuning Mistral, every government building a sovereign stack buys silicon. Nvidia's open-weights idealism is real, but it's also a demand-generation strategy. Huang wins either way, but he wins bigger in an open world.

The open-model labs: Meta, Mistral, Black Forest Labs, Arcee AI, Reflection. These companies' entire strategies depend on open releases staying legal and respectable. For Meta, open weights commoditize the model layer where it doesn't lead and protect the application layer where it does. For Mistral, Europe's flagship lab, the letter's "sovereignty" language is the whole pitch.

The platform and infrastructure layer: Microsoft, IBM, ServiceNow, Hugging Face, Replit, Box, Telnyx. Hugging Face is the open-model ecosystem's town square; its signature was inevitable. The surprise is Microsoft. OpenAI's most important partner just signed a letter that OpenAI conspicuously did not. That's not carelessness. Microsoft has spent two years hedging its OpenAI dependence, shipping its own Phi open models and hosting everyone's weights on Azure. This signature says the hedge is now policy.

The money: Andreessen Horowitz, Y Combinator, Emergence Capital. Portfolio math. A world where only three closed labs matter is a world with fewer venture outcomes. A16z has been the loudest open-source AI advocate in venture for two years; YC's batches are full of startups that exist only because open weights let them build without frontier-model pricing.

The security establishment: CrowdStrike, Palantir. The most rhetorically useful signatures on the page. The strongest argument against open weights has always been security. Having a top cyber-defense firm and the defense sector's favorite software company co-sign the claim that "defenders need access to models with comparable capabilities" preemptively blunts the national-security counterargument.

The institutional conscience: Mozilla, The Linux Foundation, American Innovators Network. These names do the historical framing work. The letter's opening analogy, that AI in 2026 is where open-source software was in the 1980s, lands differently when the Linux Foundation itself is underwriting it.

And the curiosities: Perplexity, Arena, Mariana Minerals. Perplexity builds on open models to avoid dependency on the closed labs it competes with for search queries. A mining company on an AI policy letter mostly signals how wide Nvidia cast the net.

The map: who didn't

Now the negative space. OpenAI, Anthropic, and Google, the three companies whose closed frontier models define the market, are absent. So are Amazon, Apple, and xAI.

The first three absences are the story. Every argument in the letter cuts against the closed-lab position: that concentration creates "single points of failure," that closed models "can be breached, misused, or fail in ways that outsiders cannot detect," that safety claims should rest on "real and demonstrated harms rather than assuming that closed systems are safer by default." That last line is nearly a direct shot at the safety-first licensing frameworks the closed labs have advocated.

It's worth being fair to the absent: OpenAI has shipped open-weight models, and Google's Gemma family is among the most-downloaded open models anywhere. These aren't companies that release nothing. But signing this letter would mean endorsing its policy thrust: no restrictions on open-model diffusion, distillation is fine, openness is the path to safety. That thrust is incompatible with how all three companies talk to Washington. Their absence isn't an oversight. It's a position.

The industry has now formally split into two camps, and each is asking the government for opposite things. One wants a protected frontier; the other wants an open commons. Only one set showed up with 25 signatures.

Why this matters for the rest of us

Three things to watch from here.

One: the Chinese-model question forces a decision. If Washington restricts Chinese open weights, it validates the principle that governments can gate model diffusion, a principle that could eventually reach American open models too. That's the slippery slope the letter is really about, and it's why the signatories mobilized now rather than waiting.

Two: distillation just became a lobbying position. What was a technical dispute between OpenAI and DeepSeek is now a formal policy stance backed by Microsoft, Meta, and Nvidia. Any future regulation of model training will have to reckon with 25 companies on record calling distillation a legitimate tradition of "learning from, building upon, and improving existing technologies."

Three: watch Microsoft. The most strategically loaded signature on the page belongs to the company with the deepest ties to the closed-model camp. If the OpenAI and Microsoft relationship keeps loosening, this letter will read in hindsight as one of the early public markers.

Huang closed his post by saying the world needs "both frontier closed models and frontier open models." That's the diplomatic version. The letter itself is less balanced. It's a 25-company argument that the open half of that sentence needs defending, right now, before policy forecloses it. Judging by who declined to sign, the closed half agrees that's exactly what's at stake.

Sources: the letter (PDF), Jensen Huang's post, Tom's Hardware, CNBC, Fortune, TNW