Open Weights and American AI Leadership is the letter NVIDIA CEO Jensen Huang posted on July 24, 2026 — his first-ever post on X — arguing that Washington should treat downloadable AI model weights as national infrastructure, the same way it eventually treated open-source software. Twenty-five companies signed on day one; by the next day the list had doubled to fifty, and it still hadn't picked up Amazon or Anthropic.
Short answer: Open Weights and American AI Leadership is a July 24, 2026 coalition letter led by NVIDIA's Jensen Huang, arguing the U.S. should treat open-weight AI models as strategic infrastructure, not a security risk. It grew from 25 to 50 signers in a day — Microsoft, Meta, Google, OpenAI, and Hugging Face among them. Amazon and Anthropic never signed.
I didn't take the first recap I found at face value, because the early coverage already disagreed with itself — one widely shared summary claimed OpenAI hadn't signed. When I pulled the actual PDF straight from NVIDIA's own server rather than a secondhand write-up, OpenAI was sitting right there, the 37th name in a fifty-name list. In my testing, that's the pattern with fast-moving letters like this: the primary document is a better source than whatever got written about it twelve hours later, and it's worth the two minutes to go check.
What you'll need
Nothing beyond a working link and a few minutes — this isn't a tool you install. You want the letter itself, not a summary of it, plus a way to cross-check any number someone quotes from it against a source that isn't also trying to sell you an opinion. If you want to see what "open weight" actually means in practice rather than just read about it, it also helps to have five spare minutes and either a free Hugging Face account or Ollama installed — both are covered in the step-by-step below.
Step-by-step: reading a coalition letter like this properly
1. Read the primary document first
The letter runs about 600 words and argues open weights are the AI-era equivalent of 1980s open-source software: a shared foundation that "supports most of the internet" today, in the signers' own words, and should do the same for AI. Read it before reading anyone's take on it, including this one.
2. Check the actual signature list, not the headline count
Signature counts move fast on letters like this. This one was reported at 25 companies on launch day and had doubled to 50 by the next afternoon, per Forbes — OpenAI, Google, AMD, Cisco, Cloudflare, GitHub, Block, and Ollama were among the names added in that second wave. Whatever number you see quoted, check the date it was quoted on.
3. Notice who's missing, and don't assume why
Amazon and Anthropic are both absent, which is the detail every outlet covering this flagged — Amazon is Anthropic's largest investor and a close infrastructure partner, so their shared silence reads as a decision, not an oversight. That said, absence isn't the same as opposition; neither company issued a statement rejecting the letter's arguments, they just didn't add their names.
4. Weigh the argument against who's making it
Every company on this list has something to gain from a policy environment friendly to open weights — NVIDIA sells more GPUs the more models run anywhere, Microsoft and Google host open and closed models on the same cloud, Meta's whole AI strategy is open-weight. That doesn't make the argument wrong, but it's not a neutral panel, and the letter reads differently once you know that.
5. Try an open-weight model yourself before trusting anyone's framing
The fastest way to judge whether "open weight" matters to you is to run one. My separate guide on self-hosting an open-weight model with Ollama walks through doing this in about ten minutes on a normal laptop — it's the difference between reading a policy argument and having an opinion on it.
Example prompts you can copy
Once you've read the letter, these are useful for pressure-testing it against an AI tool like ChatGPT or Claude rather than just accepting a summary:
- "Summarize the strongest argument in this letter in three sentences, then give me the strongest counterargument a closed-model company could make."
- "This letter says openness helps AI safety through transparency. What's the best evidence against that claim?"
- "List every company mentioned as a signer here and tell me what each one's business model gains from open-weight AI being treated favorably by policy."
- "I have this letter and a news article about it that disagree on a fact. Here's both — which one matches the primary source language more closely?"
- "Explain the difference between 'open weight' and 'open source' as applied to AI models, in plain English, with one example of each."
Common mistakes to avoid
The mistake I see most with a story like this is trusting the first number you see and never checking whether it was current — a 25-signer count and a 50-signer count were both accurate, just twelve hours apart. Second is treating "open weight" and "free" as the same thing; a model's weights being downloadable doesn't mean running it at scale costs nothing, and it doesn't mean the license lets you do anything you want with it. Third is assuming Anthropic's absence means it publicly opposes the letter's arguments — it didn't issue one, even after David Sacks, the White House AI adviser, said on the All-In Podcast on July 26, 2026 that the closed-lab "duopoly" wanted to "basically put a dagger through the heart of the entire American open-source ecosystem." Fourth, don't confuse a company signing a letter with that company only building open models — Microsoft, Google, and Meta all signed while continuing to ship closed frontier models too. Fifth, don't skip the primary source because a summary is faster; in my testing, at least one popular recap got a signatory wrong in a way that changed its whole framing.
Where the major labs actually stand
| Company | Signed the letter? | Has its own open-weight model | Why it signed (or didn't) |
|---|---|---|---|
| NVIDIA | Yes (organizer) | No — sells the hardware | More open models running anywhere means more GPUs sold |
| Microsoft | Yes | Phi family | Azure profits whether customers run open or closed models |
| Meta | Yes | Llama | Its AI strategy is built almost entirely on open weights |
| Yes | Gemma | Ships open Gemma alongside closed Gemini | |
| OpenAI | Yes | gpt-oss | Returned to open-weight releases after years of closed-only |
| Hugging Face | Yes | Hosts most of the open ecosystem | Open weights are the entire business |
| Amazon | No | No | Anthropic's largest investor and infrastructure partner |
| Anthropic | No | No — Claude is closed-only | Publicly cautious on distillation and open-weight risk |
The gap the letter is arguing about is real, not just rhetorical. Per Stanford HAI’s 2026 AI Index, the top closed model led the top open model by just 0.5% on the Arena leaderboard in August 2024 — by March 2026 that lead had widened back out to 3.3%, with six of the top ten Arena models now closed. Open weights caught up once and lost ground again, which is exactly the dynamic this coalition is trying to keep from becoming permanent.
Tools that make this easier
If you want to see what an open-weight model actually feels like to run rather than just read the argument for one, my open-weight AI self-hosting guide covers Ollama and the rest of the stack in about ten minutes of setup. If you'd rather stick with closed models while you decide where you land on this, my guides to ChatGPT, Claude, and Gemini all cover the no-drama way to get started, and my free AI tools roundup is the place to start if budget is the actual constraint rather than curiosity. Developers specifically weighing an open-weight coding model against a closed one should see my best AI tool for code picks. And if the theme here — checking a primary source instead of trusting the recap — is useful to you, I used the same approach on Terence Tao's Mathematics in the Age of AI lecture, where the same lesson applied.
My take
What stands out isn't the argument itself — "openness helps competition and security" is not a new claim — it's how fast the coalition assembled and how cleanly the absences map onto business interests. NVIDIA, Microsoft, Meta, Google, and OpenAI all profit from a world where open weights are politically safe; Amazon and Anthropic's shared silence maps just as cleanly onto Anthropic's closed-model business and Amazon's stake in it. Neither side is acting against its own interest, which doesn't make either side wrong. If you take one thing from this, take the habit, not the conclusion: go find the primary document before you form an opinion on what it says.
Frequently Asked Questions
Is "open weight" the same thing as free?
Not exactly. Open-weight models are usually free to download, but running them at any real scale still costs you hardware, electricity, or a hosted API bill — and the license can still restrict what you're allowed to do with the model. "Open weight" describes access to the model, not the total cost of using it.
Why didn't Anthropic sign the letter?
Anthropic hasn't published a formal statement explaining why. It's been publicly cautious about distillation and the risks of widely available open weights, and it's the only major U.S. lab that builds closed-only frontier models, which makes its absence read as consistent with that position rather than a surprise.
Is Google actually part of the coalition?
Yes, as of the current signed version of the letter — Google is listed among the roughly two dozen companies that joined the day after the initial 25, alongside OpenAI, AMD, Cisco, and GitHub, per Forbes' reporting on the expansion.
What's the easiest way to try an open-weight model myself?
Install Ollama and run a mid-size model like a 7B or 8B release directly from your terminal — no account or API key required. My open-weight AI guide walks through this in about ten minutes.
How long does it take to read the actual letter?
About three to five minutes. It's roughly 600 words plus a signature block — shorter than most of the articles written about it.