SAN FRANCISCO — The filing landed with the Securities and Exchange Commission in June, wrapped in the confidentiality that American law grants companies preparing to go public. The number inside is what has shaken the AI industry: Anthropic, the maker of the Claude models, booked more than $11.5 billion in revenue in the second quarter, roughly 14 times the $787 million it reported in the same period a year earlier.
The figure, disclosed this month as the company works through a confidential IPO filing, came with another first: positive adjusted operating income for the quarter, according to people familiar with the company’s finances. Combined with the $4.73 billion booked in the first three months of the year, Anthropic’s first-half revenue stands near $16.2 billion, putting the company on an annualized pace of about $46 billion.
The numbers arrived with the company locked in a quiet period, barred by securities rules from commenting publicly on its own financial performance. That silence has not stopped the conversation around it. The Financial Times reported that investors in Anthropic expect the company to list in October at a valuation of $2 trillion or more. If that holds, the offering would surpass SpaceX’s record June listing and become the largest initial public offering in business history.
Two people familiar with Anthropic’s finances said the company projects revenue of $190 billion to $200 billion by 2028, a figure more than four times its current annualized run rate. Bankers and investors are being asked to price the IPO against that two-year-out forecast, an unusual practice that has precedent: Cerebras Systems leaned on 2028 expectations ahead of its listing, and SpaceX priced against projections running to 2029.
The mechanics behind the revenue surge are worth parsing. Anthropic’s compute cost fell to 56 cents for every dollar of revenue in the second quarter from 71 cents in the first, a swing that analysts said reflects two multi-year compute commitments — roughly $40 billion with Google for TPU capacity and $33 billion with Amazon for AWS capacity — struck as part of funding rounds in which both companies also took equity. The two hyperscalers are Anthropic’s largest compute suppliers and its shareholders, an arrangement that has drawn scrutiny from regulators and questions from rivals about how durable the economics are.
The growth trajectory has been steep by any measure. Internal documents seen by investors show annualized revenue near $9 billion at the end of 2025, breaking $30 billion by April and surpassing $60 billion by July. In May, the company raised $65 billion at a valuation of $965 billion, with an annualized run rate of $47 billion at the time. The second-quarter print, beating the $10.9 billion the company had guided to in May, confirmed the path rather than revising it.
Morgan Stanley, Goldman Sachs and JPMorgan are working with Anthropic on the listing, according to people familiar with the matter. CFO Krishna Rao has been leading early meetings with prospective investors, keeping the conversations high-level, with no specific financial figures or valuation discussed yet, the people said.
The stakes extend beyond one company. Anthropic’s offering is shaping up as the clearest test yet of whether public markets can absorb an AI company at a trillion-dollar valuation — and whether the run-rate math that has carried the private market can survive the scrutiny of public investors. Skeptics note that the company’s first profitable quarter was achieved on an adjusted basis, with revenue heavily concentrated in cloud subsidies and enterprise deals struck at discounted rates.
Proponents point to the same data from the other side. “Two years ago the story was that every token was sold at a loss and scale only deepened the hole,” one investor familiar with the company’s books said. “The second-quarter numbers put that argument to rest. The margin coming in now exceeds the spending going out.”
Investors are also weighing what the offering means for OpenAI, which filed confidentially roughly a week after Anthropic and is preparing its own listing. The two labs have spent the past three years trading places atop the AI rankings; now their bankers are competing for the same pool of capital in what could be the busiest IPO window the technology sector has seen.
Anthropic’s own pitch, according to people familiar with the materials, rests on a total addressable market above $30 trillion — every type of work an AI model could eventually perform — larger than the $28.5 trillion SpaceX used in its record offering. The company is expected to argue that its growth rate, its enterprise distribution and its margin trajectory justify pricing ahead of current performance.
For now, the company is holding investor meetings ahead of an October window, and the quiet period keeps its executives from making the case in public. The filing that started all this sits with the SEC, waiting. If the numbers hold up to due diligence, the largest IPO in history may be priced before the year ends.
