Anthropic files publicly for IPO at valuation above $1.5 trillion

Anthropic, the maker of the Claude chatbot, has publicly filed for an IPO in the US. Shares will trade under the ticker ANT, and analysts and investors value the company, founded in 2021, at more than $1.5 trillion. The offering is positioned to rank among the largest ever, and it will test the entire AI boom: how justified are multitrillion-dollar bets on neural networks? The prospectus shows the real numbers for the first time. Revenue is growing explosively, but losses are comparable to it in scale.

Revenue grew 400-fold in three years, and so did the loss

The prospectus shows the following trajectory:

• 2023: $41 million in revenue
• 2024: $386 million
• 2025: $4.585 billion, growth of more than 1000%
• first half of 2026: $16.286 billion

In six months of 2026, the company earned more than three times what it made in all of 2025. Net loss for the half-year came to $80 billion, almost five times revenue. By the end of June, Anthropic had raised $141.1 billion, and it also took on $6.2 billion in debt to build infrastructure. Separately, the document says the company spends billions of dollars a year on compute from third-party providers.

A bet on the enterprise

Unlike OpenAI, which bet on the mass consumer market for ChatGPT, Anthropic makes its money from business. As of September 30, it has more than 14 000 enterprise customers:

• 11 900 bring in $100 thousand or more
• more than 1900 bring in $1 million or more
• about 250 bring in $10 million or more

Revenue concentration is moderate but noticeable: a quarter comes from the five largest customers, though none accounts for more than 10%. Losing a couple of them would visibly hurt the numbers, and the company says so directly in its risk factors. In the prospectus, management calls the pace of adoption confirmation of Claude's position at the frontier of AI. It describes agents as a "country of geniuses in a data center" available to every organization and every sector of the economy.

Ninety pages of risks

The risk section runs to more than 90 pages. Anthropic identifies four categories of AI threats: misuse by malicious actors, loss of control, authoritarian use and economic disruption. The company acknowledges that the most powerful models can give meaningful help to those developing chemical, biological and other dangerous weapons. It calls its technical safeguards incomplete: they may become outdated, and they have been bypassed before and may be bypassed again.

The document gives a concrete example: in September 2025, employees recorded what they assess as the first large-scale cyberattack carried out with almost no human involvement. A hacking group linked to the Chinese state manipulated the Claude Code tool and tried to breach roughly 30 targets worldwide, including technology companies, banks, chemical plants and government agencies.

There is also a political front. The administration of Donald Trump is in conflict with Anthropic over the use of its products for military purposes. The result was the company's designation as a supply chain risk, the first time this has happened to an American firm. No military agency can now sign contracts with it. Anthropic is contesting the decision. There is also a lawsuit alleging that 274 Elvis Presley songs were used to train the models.

The story has a paradox. Dario Amodei, the co-founder who left OpenAI, has publicly called for slowing AI development because of existential risk. The listing, though, could bring a wave of money into the company and the sector, which would speed up the race even more.

What to watch before the offering

For investors, the key question is the margin trajectory: can revenue, growing by billions per quarter, catch up with compute costs? For enterprise customers, two factors matter. The first is the stability of a vendor that burns tens of billions in half a year but has received more than $141 billion in funding.

The second is legal risk: the Pentagon blacklisting and the copyright lawsuits could affect contract terms. Anyone building products on Claude should check in advance how dependent they are on a single model and build in a fallback. The first behavior of ANT shares will also be telling: it will show whether the market is ready to pay a $1.

5 trillion valuation for an AI leader with a loss this large.