On Monday, June 1, Anthropic confidentially submitted its draft S-1 to the Securities and Exchange Commission, setting the stage for an IPO that would value the Claude maker near $965 billion. The filing comes weeks after a $65 billion Series H round and lands the company ahead of OpenAI’s most recent $852 billion mark. A $1.25 billion monthly GPU commitment and a $47 billion annualized revenue run-rate give analysts something to chew on. But the number that matters most isn’t in any spreadsheet.

It’s the gap between what Anthropic promised and what it now has to deliver.

For years, the company’s entire pitch to talent, to policymakers, and to the public has been that it takes AI safety seriously. Not as a compliance checkbox. Not as a marketing angle. As the organizing principle of the firm. The corporate structure—a public benefit corporation with a long-term benefit trust that can override the board on safety grounds—was designed to prove that profit and precaution could coexist. Now the public markets get to decide what that structure is actually worth.

The Safety Premium Nobody Knows How to Price

Public markets are mechanism for pricing the predictable. Revenue growth, gross margins, customer concentration, capex trajectories—these are legible. But Anthropic is asking institutional investors to price something fuzzier: the probability that being careful now prevents a catastrophe later. That’s not a line item. It’s theology.

“The roadshow decks are going to be fascinating,” one institutional allocator told me over a drink at a Midtown conference last week. “They’re selling Claude, obviously. But the real product is ‘we won’t kill you.’ How do you put a multiple on that?”

The cynical answer, and the one that will circulate widely this week, is that you don’t. That the safety narrative was always a recruiting wedge and a regulatory charm offensive, and that once the quarterly earnings calls start, the trust board will discover exactly how much power it has when it tries to slow a product launch that the Street expects. The cynical answer has the virtue of being easy to tell and hard to disprove.

But the cynical answer also misses something.

What the Public Benefit Structure Actually Does

The long-term benefit trust isn’t a suggestion. It has the power to remove and appoint board members if it determines the company is straying from its safety mission. That’s real governance, not a press release. And it means Anthropic’s IPO is, among other things, a natural experiment in whether a fiduciary structure designed to constrain profit-seeking can survive contact with shareholders who bought in precisely because the profit potential looked too big to ignore.

Anthropic raised $65 billion privately by convincing investors that safety was a moat—that enterprises and governments would pay a premium for an AI provider they could trust not to blow up in their faces, reputationally or literally. The public listing tests whether that logic scales. If it does, the trust board becomes an asset: the thing that justifies a higher multiple than OpenAI’s purely for-profit structure. If it doesn’t, the trust board becomes a target.

“Every shareholder who buys Anthropic at a $965 billion valuation is implicitly long the idea that safety creates durable competitive advantage,” a former SEC enforcement attorney who now advises tech firms on governance told me in a courthouse hallway during a break in an unrelated hearing. “The moment that thesis cracks, the lawsuits start. And they’ll name the trust board personally.”

The Real Tension Isn’t Profit vs. Safety

It’s transparency vs. control. Confidential S-1 filings let companies work through SEC comments before the public sees the financials. That’s normal. But Anthropic’s entire identity rests on claims about internal processes—alignment research, constitutional AI, capability evaluations—that are largely unobservable from outside. The S-1 process will force some of this into view, but only some. Revenue figures, customer lists, risk factors. Not testing protocols. Not the minutes of trust board meetings where a product launch was delayed over safety concerns. Not the near-misses.

The company can’t have it both ways forever. If it wants a safety premium in its stock price, it eventually has to let investors see the safety work. But the safety work, by its nature, involves disclosing things that look alarming out of context—capabilities you discovered and contained, deployment risks you averted, scenarios that came closer than anyone would like. Public companies are notoriously bad at nuanced disclosure. They’re good at earnings beats.

Anthropic’s bet is that the market will reward a company that says: we’re going to move a little slower, disclose a little less, and ask you to trust a governance structure you’ve never seen before. The alternative—the OpenAI path—is to race ahead and argue that speed is its own kind of safety.

Monday’s filing doesn’t settle that argument. It just moves it to a much bigger room, with much less patience.

Sources