Anthropic has submitted a confidential registration statement to the Securities and Exchange Commission for an initial public offering, according to people familiar with the matter, placing the most closely watched balance sheet in artificial intelligence before public investors at a moment when the industry’s pricing power is under question.
The company, founded in 2021 by Dario and Daniela Amodei after they left OpenAI, is seeking a valuation of as much as $965 billion, the people said. Revenue is running at an annualized rate of roughly $47 billion, a figure that would put the listing among the largest technology IPOs in recent memory and give the market its first public price tag for a frontier AI laboratory.
Anthropic declined to comment. Under the confidential process, financial details remain sealed while regulators review the paperwork, and the terms of the deal will not become visible until the company files its full registration statement, a step that typically comes within weeks or months. That document will give investors their first detailed look at the cost structure of a business built largely on rented computing power.
The stakes extend well beyond one company. CNBC, citing analysts, said gross margin will be the metric that either validates or overturns the AI growth story. If Anthropic can demonstrate that its premium-priced access to the Claude models is sustainable, the argument runs, the entire frontier AI group gains; if margins come in below expectations, the result could be a chain of valuation corrections across the sector.
The arithmetic explains the attention. Anthropic sells most of its output through API tokens and enterprise subscriptions, and its two largest bills are compute contracts with Amazon and Google, both of which are also investors. Computing absorbs the majority of each revenue dollar, so Wall Street will scrutinize the spread between what the company charges for tokens and what it pays for the machines that produce them.
The comparison investors will reach for is with software companies that sell at high margins and hardware companies that do not. Anthropic sits in between: its product is software, but its cost base looks closer to a utility’s. The company has argued that scale will push margins up as models become more efficient and compute prices fall, a claim that cannot be tested until the numbers are public. The registration statement will show whether that argument has a floor under it.
The filing is the first genuine stress test of the 2026 wave of AI listings, a queue that has been building for months. OpenAI has been reported to be preparing its own offering at a valuation above $850 billion, and the White House has discussed taking a government equity stake before any listing. Each deal will be priced against the others, which raises the pressure on underwriters to get the numbers right the first time.
A quieter threat sits on the demand side. Corporate customers are adopting so-called model routing software that automatically sends simple requests to low-cost models and reserves frontier systems for hard problems. The practice cuts the premium volumes on which Anthropic’s pricing depends and gives procurement teams a visible way to reduce token bills. CNBC has reported that the trend is spreading through enterprise accounts and that it complicates the assumptions behind frontier AI valuations.
For the Amodeis, the filing caps a rapid rise. The company was founded on the belief that safety research and commercial ambition could coexist, and it was organized from the start as a public benefit corporation, a structure meant to keep long-term judgment ahead of short-term profit. Its Claude models have won a following among developers and corporate buyers who pay for reliability rather than spectacle, and successive generations have pushed the company into the front rank of the industry.
Amazon and Google have committed billions of dollars to Anthropic, and the company has become one of the most richly valued private businesses in the world. The two cloud giants hold a dual position as investors and landlords, a relationship that regulators and public investors have already begun to question. How the IPO prices that arrangement, and what it means for compute costs, will be among the most parsed details of the filing.
What remains unknown is whether public investors share the enthusiasm at the numbers under discussion. The confidential process gives Anthropic time to wait for a favorable window; the market will not wait forever. If the IPO prices near the top of the range, it will be read as confirmation that frontier labs can be durable businesses. If it prices below, the discount will echo through every AI company standing in line behind it.
Analysts said the outcome will shape how OpenAI approaches its own listing and how quickly other AI companies move. A strong debut would encourage copycat filings; a weak one would slow the pipeline. Either way, the offering will produce the first hard number the market can use to test the claims of the industry’s most valuable private companies.
For investors, the coming weeks are a study in patience. The registration statement, when it appears, will show what frontier AI actually earns, what it spends, and whether the gross margin question has an answer. Until then, the price talk remains talk.
