On Friday, July 24, Nvidia published an open letter urging Washington not to restrict open-weight AI models. Twenty-five companies signed. By Sunday, July 26, the list had doubled to fifty — OpenAI and Google among the new arrivals. Two names remained conspicuously absent: Amazon and Anthropic.

Anthropic’s absence was not an oversight. The company had just published its own document, “Our Position on Open-Weights Models,” laying out the case that releasing powerful model weights creates irreversible safety risks. The timing was not coincidental. Neither, I suspect, was the audience.

The conventional reading of this standoff is that it represents a genuine philosophical divide in the AI industry — safety-conscious labs versus open-science advocates, caution versus acceleration. That framing is not wrong, exactly. But it misses something more immediate and more revealing. Anthropic is reportedly preparing for a massive initial public offering. And nothing concentrates the corporate mind like a roadshow.

The Real Audience Isn’t in Washington

Read Anthropic’s position paper alongside the Nvidia letter and the contrast is instructive. The Nvidia document is a classic piece of D.C. lobbying — short on technical detail, long on economic competitiveness, signed by a coalition that spans the supply chain from chip designers to cloud providers. It is aimed at the Commerce Department and the National Security Council.

Anthropic’s paper is different. It is longer, more technical, and structured like a risk assessment. It names specific threat models — bioweapons proliferation, autonomous cyberattacks — and argues that open-weight releases make those risks uncontainable. It reads, in other words, like a document designed to be cited in an S-1 filing.

One institutional allocator, forwarding the position paper to his firm’s research distribution list Monday morning, added a single line above the link: “This is the moat slide.”

He was not being cynical. He was being accurate. For a company about to ask public markets to value its intellectual property, the worst possible signal would be ambiguity about whether that IP will remain proprietary. Anthropic just removed the ambiguity.

The Google Tell

The most revealing detail in the Nvidia letter saga is not who signed but who signed after the first round. Google, an Anthropic backer, joined the second tranche. Amazon, Anthropic’s largest investor and the company on whose Trainium silicon Anthropic trains its models, did not.

This is not a coincidence of scheduling. It is a disclosure of alignment. Amazon and Anthropic are structurally intertwined in a way that Google and Anthropic are not. Amazon’s cloud business benefits directly from a world in which frontier models remain closed, proprietary, and served through APIs that run on Amazon hardware. Google’s cloud business can afford to be more ecumenical.

The split tells you that the open-weights debate is not really a debate about safety philosophy. It is a debate about business models. And Anthropic’s business model, as of this month, is about to include quarterly earnings calls.

What the Safety Argument Leaves Out

None of this is to say that Anthropic’s safety concerns are insincere. The company has been consistent on this point for years, and the specific risks it identifies — particularly around biological misuse — are not imaginary. But consistency is not the same thing as purity of motive, and the IPO timeline adds a dimension that the position paper does not acknowledge.

A company that plans to remain private can afford to take positions that look principled but cost money. A company that is about to be valued by public markets cannot. Every position it takes will be read through the lens of discounted cash flows. The open-weights paper, whatever its intellectual merits, is also a promise to future shareholders: we will not give away what you are about to pay for.

The irony is that this makes Anthropic’s position more credible, not less. A safety commitment that happens to align with a business model is more durable than one that requires constant self-denial. Wall Street understands this. It is the AI ethics community that seems uncomfortable with the implication — that the strongest guarantor of caution might be, of all things, the profit motive.

The Letter Is a Lobbying Document. So Is the Paper.

The Nvidia letter has been covered, correctly, as a lobbying effort. Fifty companies do not coordinate on a weekend press push without someone tracking legislative language. But Anthropic’s position paper is also a form of lobbying — aimed not at the Commerce Department but at the institutional investors who will decide whether the IPO prices at the top or the bottom of the range.

The difference is that one form of lobbying is legible as lobbying and the other looks like principle. The market, to its credit, can tell the difference. The question is whether the rest of us can.

Sources