Anthropic, the big artificial intelligence lab, wants everyone to understand that it’s deeply devoted to developing advanced AI “for the benefit of humanity.”
We know this because it has placed this principle front-and-center in the preliminary prospectus for its upcoming initial public stock offering, which was made public on Monday. It’s what Anthropic defines as its “Company Purpose.” The term, always capitalized, appears 42 times in the 318-page document, so it must mean something important.
On the other hand, Anthropic needs you to know that its work may not actually turn out to be boon for humanity. “We cannot assure you that we will achieve our Company Purpose or that the expected positive impact…will be realized,” it acknowledges in the prospectus.
Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.
— Torsten Slok, Apollo Global Management
That’s not all that the company is asking prospective investors to take on faith. Its financial success is dependent on its revenues handily outpacing its expenses, which already run into tens of billions on an annualized basis, mostly spent on acquiring chips and computing resources needed to build out its AI ambitions.
Common shareholders will have to have faith that Anthropic’s seven founders will always do the right thing for humanity, because the share ownership is set up so that public shareholders will have zero voice in how the company is run. More on that in a moment.
By almost any measure, the Anthropic S-1 (to use the official designation for initial prospectuses) is an extraordinary document. Market expectations are that the company could raise about $100 billion through the stock issue, with its total valuation estimated as high as $2 trillion.
That would place Anthropic ahead of the most recent mega-stock issue, which was the SpaceX IPO in June, which raised about $86 billion and valued that company at about $1.75 trillion. The Anthropic IPO is expected to launch sometime next month.
Another remarkable feature of the S-1 is the effort the company makes to warn investors that it will be playing with fire to make its billions and that the future of humanity hangs in the balance, and not in a good way.
Among the nearly 100 pages of “risk factors” cited in the prospectus are that its work could lead to the extinction of humanity — specifically, that “sufficiently advanced AI systems could pose catastrophic or existential risks to society if not developed and deployed responsibly.”
Anthropic says it does all it can “to mitigate such risks, but there can be no assurance that our efforts will be successful,” which might be less than comforting to critics of AI and the humans who are developing the technology.
It’s not unusual, of course, for IPO prospectuses to carry a fair helping of conjecture and aspiration. Anthropic’s S-1 takes these features perhaps a bit further than the traditional S-1, but qualitatively it’s a fair example of the genre.
So let’s look closer at what Anthropic has to say for itself. (I asked Anthropic for comment on the following issues, but the company declined, citing the quiet-period rules federal regulators impose on companies during the IPO process.)
In strictly financial terms, the company’s progress has been admirable, at least as it concerns the top line. The prospectus reports that revenue for the first six months of this year grew to $16.3 billion, up from $1.2 billion in the same period a year ago. Operating expenses, however, also grew rapidly, particularly for computing and infrastructure — nearly $9 billion in the first half of 2026, versus $2.3 billion a year earlier.
The result was an operating loss of $693 million in the first half, narrowed from the $2.1-billion loss a year earlier. (Using generally accepted accounting standards, its first-half loss was actually $80.7 billion, but $72.4 billion of that was the result of an accounting provision for convertible notes the company had issued to cover start-up expenses.)
That brings us to the expenses the company will face in the future. Here the picture turns darker.
In part because of the intense competition for computing resources in the AI sector, Anthropic has had to commit to several non-cancellable deals for computing chips and data center capacity, meaning it will be on the hook for those payments whether it uses the products or not.
Depending on how one counts, they total roughly $500 billion. That includes $110 billion payable to Amazon, $114.5 billion to Google, $44.9 billion to Microsoft and $120.5 billion to SpaceX.AI, a subsidiary of SpaceX.
Some of the commitments run for 10 years. They represent sizable and largely inescapable overhead, raising the prospect of financial stringency down the road.
What if Anthropic’s customer growth dwindles or stagnates? That’s a troubling prospect, given the extent to which its rich valuation rests on expectations that its recent pace of growth will continue into the limitless future.
Already, corporate customers of AI firms are beginning to cut back on their consumption as fees for their services rise and doubts about their value proliferate.
“Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech,” Torsten Slok, the chief economist at the investment firm Apollo Global Management, wrote last week.
If non-tech companies pull back on AI, the consequences for firms such as Anthropic could be dire. In its S-1, indeed, Anthropic acknowledged that “a substantial portion of our revenue is driven by a limited number of customers”: Its five biggest customers provided about 25% of its revenue in the first nine months of this year, though none represented more than 10%.
Many customers, Anthropic said, aren’t bound by long-term contracts “and are able to reduce or .. cease spending at any time and for any reason.” If competition among AI companies intensifies, watch for financial pressure on Anthropic to mount.
That brings us to the question of what shareholders could do if Anthropic’s management proves unequal to the challenge. The answer is: Nothing.
That’s because of the unique five-class share structure the company has established. The Class A common shares, which are what will trade on Nasdaq upon the IPO, have one vote per share, but won’t be receiving dividends “in the foreseeable future,” according to the prospectus.
More important, at no time will the Class A holders have voting control of Anthropic, as the class will never have more than 49.9% of the vote on corporate policies or the election of directors. That’s because the company’s single Class F share will always have 50.1% of voting power. That share will be controlled jointly by Anthropic’s seven co-founders, including Chief Executive Dario Amodei and his sister Daniela, the president and chair of the board. That arrangement will continue until about three years after the number of co-founders remaining with the company dwindles to fewer than three.
Among the other share classes is Class T, a single share given to the company’s independent Long-Term Benefit Trust, which is overseen by experts in AI safety and public policy and has the authority to elect a majority of Anthropic’s board.
The bottom line is that the control of the Amodeis and their co-founders is designed to be impregnable. In their hands rests the Company Purpose. Should that be comforting? The best one can say is: Maybe. Among the murkier aspects of this setup is what will happen if the trustees come to feel that the co-founders have lost their devotion to safety and their board majority tries to right the ship.
The Company Purpose sounds solid, but is it? It evokes what was said to be the mission statement of Google’s founders at the time of its IPO, which was “Don’t be evil.” That was a righteous-sounding if vague principle that turned out to be largely wishful.
In the fullness of time it was arguably subjugated by the burgeoning corporation’s commercial impulses. By 2011, a mere seven years after Google’s IPO, the Federal Trade Commission charged that it had deceptively breached consumers’ privacy, and a year later charged that Google had breached the privacy of users of Apple’s Safari internet browser. Google settled both cases without admitting wrongdoing, but paid what was then a record $22.5-million civil penalty the second case. Google also has come under fire for feeding users of its YouTube video service extremist content. Google has responded by trying to tweak YouTube’s recommendation algorithms to suppress such content.
That’s not to say that Anthropic will go down the same path, only that writing worthy intentions into an official document isn’t the same as writing them into stone. In other words, we can admire Anthropic’s goals, but that doesn’t mean we have to bank on them.
