50 AI companies signed Huang's open-weights letter in a day. Amazon and Anthropic didn't.

Jensen Huang's first post on X came on July 24, 2026, and it wasn't a product launch or a keynote teaser. It was a policy letter titled "Open Weights and American AI Leadership," and by the next afternoon the signature list had doubled from 25 to 50, with a second wave still arriving while reporters were still filing their first version. The post cleared 11 million views. The document was the easy part. Distribution was the event.

The letter asks Washington not to restrict the release of "open weights," the trained parameters of AI models that anyone can download and run. The position is that keeping American AI competitive means shipping those models freely, the same way the open-source software stack already does. Chipmakers, cloud providers, security vendors, app companies, and the venture firms behind them are now visibly on that side. The second-wave signatories include OpenAI, Google, AMD, Cisco, Cloudflare, GitHub, Block, and Ollama, a cross-stack lineup that runs from silicon to apps to the infrastructure in between.

Two companies are absent from every version of the letter as of July 25, 2026: Amazon and Anthropic. That pairing is the actual industry signal.

Amazon is Anthropic's largest investor, and Anthropic trains and serves models on Amazon's Trainium chips alongside other hardware. Google is also an Anthropic backer, but Google signed anyway, which isolates the Amazon–Anthropic pairing as the more specific absence. Both companies sell closed frontier access, Anthropic through its API and Amazon through the Bedrock distribution layer, and both have a financial stake in weights that stay hard to copy. That does not prove motive. Neither has explained, and the Forbes analysis by Sandy Carter that surfaced the doubling is explicit that the two readings can be true at once: a commercial interest in keeping frontier access gated, and a substantive safety position that released weights cannot be recalled.

The safety argument is the one worth holding. Anthropic has long argued that once model weights are out, they are out. There is no patch, no recall button, no way to claw them back from a misuse case. That is a legitimate critique of the open-weights policy posture, and it does not soften into a marketing line if you take it seriously. It also does not undercut the alignment on the other side of the letter. The signatories are not arguing that every model should be open. They are arguing that the default should be open, and that the closed-frontier business model should not be the only one Washington protects.

The tempo is what changed the politics. Roughly two dozen companies joined after publication, and two hosted copies of the letter carried different signature counts on the same afternoon, because the names were still being added. That sequence is the new shape of AI policy: a position becomes visible, distribution does the work, and the no-shows draw the fault line. The wire will report the signature count. It will not capture the mechanism, that Huang's first X post was the trigger, that the second wave followed attention rather than deliberation, and that the alignment now spans the parts of the stack that actually ship product.

A separate signal sits next to the no-shows. Per the same Forbes column, Anthropic "just took over the #1 position in the Enterprise." The ranking itself is not named in the visible source text, so treat that as a labeled data point rather than a settled industry claim. The direction is what matters: the company most associated with the closed-frontier counterargument is also the one gaining the most enterprise ground, which makes the open-vs-closed frontier-AI question a live business argument, not just a policy preference.

OpenAI and Google signed, even though both sell closed frontier access of their own. Amazon and Anthropic didn't. The 50 signatures on the letter are not the coalition. The two absences are.