For most of 2026, a big question in the AI industry has been when OpenAI and Anthropic would go public. For Anthropic, the process is now visibly underway. The company has submitted a registration statement on Form S-1 in the United States, and the document opens up its finances in a way few AI companies have ever allowed.
The figures below come from a report by the Spanish tech outlet Xataka, which summarizes the filing. I have not read the S-1 itself, so treat the numbers as reported rather than independently verified.
What an S-1 is, and why it matters
An S-1 is the paperwork a company must file with the SEC, the US securities regulator, before its shares can trade on a US exchange. It lays out the business, the accounts and the risks. Anthropic had already submitted a confidential draft on June 1, so the public filing was expected, and some numbers had leaked in recent months.
The filing does not say when the listing will happen. According to Xataka, the official announcement is silent on timing.
The headline numbers
The reported revenue growth is dramatic:
- First half of 2026: about $16.3 billion in revenue, versus roughly $1.16 billion in the same period of 2025, around 14 times more.
- Full year 2025: about $4.6 billion.
- Full year 2024: about $386 million.
The losses are just as eye-catching. Net losses for the first half of 2026 were reported at about $80.7 billion, up from about $12.9 billion a year earlier. Fortune, as cited by Xataka, describes this as one of the largest half-year losses ever recorded by a company.
Why the loss looks bigger than it is
Most of that net loss comes from a single accounting item. About $72.4 billion of it relates to a SAFE, a contract in which investors put in money in exchange for future shares. Take that out of the picture and the operating loss for the half was about $693 million, compared with about $2.1 billion in the first half of 2025.
That does not make the business cheap to run, and the source does not explain the accounting mechanics. But it shows why the headline loss and the operating picture tell very different stories.
Where the money goes
The spending breakdown for the first half of 2026 shows what building frontier AI costs:
- Compute and infrastructure: about $8.9 billion, up from about $2.3 billion a year earlier.
- Sales, marketing and partnerships: about $3.2 billion, versus roughly $120 million before.
- Personnel: about $3.2 billion, including about $2.2 billion in stock-based pay. Xataka cites a comparison figure of $538 million for stock compensation in 2025.
Computing power is the largest single line. That is a useful reminder for anyone using AI tools: the free or low-priced chat window sits on top of very expensive infrastructure.
Who is paying
On the income side, the filing reportedly shows nearly 12,000 customers that spent $100,000 or more over the last twelve months. Investing, cited by Xataka, adds that more than 1,900 of them exceed $1 million and about 250 exceed $10 million. Revenue per employee is reported to have risen from about $1.4 million in 2025 to $6.6 million in 2026.
These are business-heavy numbers. They suggest that Anthropic’s growth is driven largely by companies rather than casual consumers, though the source does not give a full breakdown by customer type.
The risks the company itself flags
An S-1 is also where a company must spell out what could go wrong. According to the report, Anthropic warns of several things:
- Past growth does not guarantee future growth.
- Results have depended on a small number of customers and distribution channels.
- Some cloud providers are at once customers, suppliers and investors, the kind of overlap often called circular financing in AI.
- AI models may act autonomously and take harmful actions nobody ordered.
On that last point, the filing notes that Anthropic has delayed some launches out of caution. The example given is Mythos, which is limited to a cybersecurity effort called Project Glasswing. The report’s author urges some care with the autonomy language, since AI systems are software, not living things, and the passage is partly a transparency requirement.
The company’s stated pillars
The filing also describes three pillars: focus, talent and research. In short, Anthropic says it concentrates talent and computing investment on the highest-value work, attracts people motivated by its mission, and integrates capability and safety research so that both shape how models are designed, trained and deployed.
The author singles out the research point as the most concrete. Anthropic has concentrated on fewer products and models while competitors tried to cover everything, and the company argues that building on earlier, mature models lets customers proceed with confidence.
How to read news like this: a quick checklist
- Separate revenue from profit. Fast growth and large losses can coexist, especially when compute costs climb alongside demand.
- Look for one-off or accounting items. Here, one SAFE-related item accounts for most of the net loss.
- Check concentration. Dependence on a few customers or partners is a risk, even when the totals look great.
- Read the risk section. It is often the most candid part of a filing, though it is written with legal caution in mind.
- Check the source. If you are reading a summary, go to the SEC filing for the exact wording before relying on any figure.
None of this is investment advice. A filing is a snapshot, and the numbers here are from a secondary report.
Summary
Anthropic’s S-1 gives a rare look at the economics of a frontier AI lab: revenue reportedly up about 14 times year over year in the first half of 2026, enormous spending on computing, and a headline net loss largely driven by one SAFE-related accounting item. The company also admits to risks around customer concentration, circular relationships with cloud providers and autonomous model behavior. When the listing will happen remains unclear.
