Dario Amodei has spent years describing artificial intelligence in the language of civilization, science and God. In private meetings with prospective investors this summer, the Anthropic chief executive has been talking about something more mundane: revenue.
The company is working to steady investor confidence before a public listing that could rival SpaceX in scale, according to people familiar with the matter. The Wall Street Journal reported Monday that Anthropic has intensified its courtship of investors, and the New York Post described a charm offensive by the company’s leadership. The campaign reflects a gap that has grown harder to ignore: the distance between Amodei’s grand narrative about artificial general intelligence and the company’s actual financial statements.
Anthropic has become one of the two most valuable private AI labs, backed by Amazon and Google and known for the Claude chatbot and its safety-first brand. Revenue has grown quickly, and the company sells access to its models through application programming interfaces and enterprise subscriptions, a business that rises and falls with corporate AI budgets. Investors who have heard the pitch say the sums still look small next to the valuations being floated for the listing, and that the story has long run ahead of the numbers.
Anthropic has raised large rounds at valuations that climbed steeply over three years, and its cap table includes some of the same institutions that will be asked to buy again in the IPO. The company has raised more money than most startups ever will, yet its operating costs are enormous: frontier models consume compute at a scale that keeps the business far from consistent profitability. That combination — huge capital needs, huge compute bills, huge ambitions — is what makes the IPO such a test of the firm’s narrative.
Recent months added new questions. Anthropic’s latest flagship model, Fable 5, drew criticism after a controversy over errors in its biosafety screening, in which the system wrongly blocked a stream of benign requests, according to people who reviewed the incidents. The company also faces copyright litigation over the data used to train its models, a case that could shape how much of its technology stack is exposed. Each episode has given skeptical investors a concrete reason to discount the narrative, and Amodei’s own writings have done little to narrow the debate: he has published long essays describing a future in which AI accelerates scientific discovery and reshapes entire fields, a vision that inspires some backers and unsettles others.
The IPO planning has proceeded on parallel tracks. People familiar with the matter said the offering could be among the largest in technology, on the order of the listing SpaceX has been preparing, and that banks have been sounding out demand. Inside the company, executives have described the same period as a chance to show investors that the safety-first positioning that defined Anthropic is not a cost center but a reason customers pay a premium.
The contrast with OpenAI sharpens the conversation. OpenAI’s ChatGPT has become a consumer phenomenon, with the company racing through product releases and infrastructure deals as it prepares its own listing. Anthropic’s Claude is more of a workhorse: deployed by developers, embedded in enterprise software, measured in tokens rather than downloads. Investors who compare the two see one company selling to the world and another selling to companies, and they want to know which model earns more per dollar of compute.
Amodei’s answer, delivered in essays and investor meetings alike, is that the future is closer than it looks and that Anthropic’s restraint is a feature, not a bug. He has argued that the company’s careful approach to safety will protect it from the regulatory and reputational damage that could hit faster movers, and that the enterprise base built in the meantime will compound. The pitch has won over a long list of marquee backers, including some who missed the early rounds of OpenAI and are reluctant to repeat the mistake.
Not everyone is convinced. Some investors describe the company’s valuation story as having shifted from faith to earnings: the era in which a slide deck about AGI could carry almost any price tag is over, and the people being courted for the IPO say they want profit-and-loss statements, unit economics and a path to cash generation, not just a timeline for superintelligence. The meetings themselves have been small and pointed, according to people familiar with them, with questions returning again and again to margins, compute costs and the durability of demand.
The charm offensive will face its first real test when the company files its prospectus. The document will put numbers next to the narrative for the first time, and the gap between the two will determine the size of the offering. Amodei can still describe the future in grand terms; the paperwork will describe the present in decimals. Investors are asking for both, and they are in no hurry to pay for the first without the second.
