The developer of Claude has embarked on a path to a stock market debut: on October 26, 2026, Anthropic filed a Form S-1 registration statement with the US SEC for an IPO of Class A shares, with a planned listing on Nasdaq under the ticker ANT. Along with the filing, the market saw the behind-the-scenes economics of the frontier lab—from semi-annual revenue and loss structure to computing power commitments worth hundreds of billions of dollars.

image
image
image

What happened

Anthropic, PBC filed a Form S-1 registration statement with the US SEC for a proposed IPO of Class A shares. The listing is planned for the Nasdaq Global Select Market under the ticker ANT, with Goldman Sachs and Morgan Stanley named as underwriters. The number of shares, price range, and company valuation have not yet been announced, and the sale of securities will begin only after the statement becomes effective. According to the S-1, revenue for the first half of 2026 was $16.3 billion—more than for the entire year of 2025, when it was $4.6 billion. The net loss for the same period reached $80.7 billion, primarily due to non-cash accruals and AI infrastructure costs; however, in the second quarter, the company achieved operating profit for the first time, amounting to $974 million. The filing also discloses fixed computing power commitments of approximately $430 billion, including contracts with Broadcom, Google, Amazon, and Microsoft.

Context

Anthropic is one of the leading developers of frontier models and the creator of Claude, having remained a private company until now. Its valuation reached $965 billion in May 2026, with expectations for the final offering around $2 trillion, significantly higher than OpenAI's $852 billion. Until now, the economics of frontier labs could only be judged by venture round estimates and leaks, so detailed public disclosure of financial reporting changes the very nature of the data available for sector analysis. The filing shows what Anthropic's business is built on: 80% of revenue comes from token consumption and 20% from subscriptions, meaning income is almost directly tied to the volume of available computing power.

Why this matters for the industry

The disclosed figures become a public benchmark for the cost of large-scale AI production: for the first time, not only revenue is visible, but also the cost structure, including approximately $9 billion in AI infrastructure costs for the half-year, and after the listing, such data will be published quarterly. Since the majority of revenue is linked to token consumption, the API economy of agents and automation built on Claude is the company's main business, not a byproduct, setting a benchmark for product builders. The ratio of computing commitments to current revenue is the most indicative metric for scaling researchers: it fixes the real cost of betting on exponential growth. If token revenue grows faster than computing costs, the scaling hypothesis is confirmed; if not, the commitments will become a public case of overheated computing leverage. For startups and corporate clients, the execution of these contracts means a predictable infrastructure horizon for years to come. There is also a limiting factor: about 20% of revenue comes from just two clients, so current growth cannot be extrapolated to the entire market.

Why this matters for users

If the offering takes place, private investors will for the first time have the opportunity to directly buy shares in the company creating Claude. On the day the S-1 was filed, nothing practical changes: Claude's products, API, prices, and latencies remain the same, as this is a legal step before a stock market debut, not a model release. The disclosed data on revenue, losses, and computing costs is a rare transparent slice of the AI business economy, useful for both investors and those building AI products. It is worth watching the preliminary prospectus, where the price and volume of the offering will appear, as well as the unusual lockup schedule: some employees will be able to sell their shares on the first trading day.

What is still unknown / limitations

The S-1 does not disclose the volume of available computing capacity, its utilization, or the cost of inference per million tokens, so any conclusions about the sustainability of the service and API infrastructure remain interpretations rather than verifiable facts. High customer concentration prevents considering revenue growth as proof of broad market demand—data on customer cohorts from future reports is needed. The sustainability of the second quarter's operating profit and the full cash breakdown of the loss will only become clear from the first quarterly reports of the public company. Finally, the statement says nothing about models, architectures, or benchmarks, so Claude's capabilities cannot be assessed based on this document.

Sources

Author

Look at AI, editorial team