$80.7 million. That's what Anthropic lost in the first half of 2026, against more than $16 billion in revenue. Over the same months a year earlier, the red ink ran to $13 billion.
The figures come from the S-1 the company filed Monday, which sets the Anthropic IPO in motion for real. According to WIRED, the stock is set to list on Nasdaq under the ticker ANT. The Wall Street Journal has suggested a debut as soon as later in November.
Profitable? Not yet. But a turnaround this steep inside twelve months is exactly what makes investors reach for the highlighter.
The numbers behind the Anthropic IPO
Revenue is the headline. Sales for the first six months of 2026 topped $16 billion, compared with roughly $4.5 billion for all of 2025. Do that arithmetic slowly. A single half-year has already left the previous full year far behind.
Who is paying? Mostly big spenders. Across the trailing twelve months, upward of 11,900 customers each paid Anthropic at least six figures. About 1,900 of them cleared $1 million, and roughly 250 went past $10 million. This is a company funded by businesses running Claude at scale, not by consumers idly testing a chatbot.
The prospectus skips the part everyone is waiting for. It names no fundraising target and no valuation. In August, the Times reported that Anthropic was contemplating a raise above $100 billion, priced against a company worth around $2 trillion. Until an amended filing shows a price range, treat that as chatter rather than a plan.
A risk section that reads like a confession
Every S-1 carries risk factors. Lawyers insist. Anthropic's version is unusually vivid. It begins by telling buyers the stock is "a high degree of risk," then maps out how things might fall apart.
Some worries are ordinary: securing enough computing power, staying ahead of rival labs. Others are anything but. The company says advanced AI could pose catastrophic or even existential danger to humanity. It sketches scenarios in which hackers use the technology to launch cyberattacks or engineer novel biological weapons, and in which authoritarian regimes use it to surveil their own people. It also raises the possibility that Anthropic loses its grip on the models it builds.
I've read plenty of boilerplate, and this isn't quite that. A firm that has spent five years telling the public AI is dangerous must now tell its shareholders the same. The consistency is almost endearing. Will public-market buyers treat it as a feature or a flaw? Different question, and nobody has the answer yet.
Washington is a line item too
The filing also names one very specific headache: the Trump administration labeled Anthropic a supply-chain risk. The company is fighting that label in court and admits it can't predict how the cases will resolve. A loss, it warns, could damage sales to government agencies, as well as to private firms that work with the federal government.
That second group matters more than it sounds. Plenty of large enterprise buyers hold federal contracts, and they tend to steer clear of vendors that complicate those relationships.
The OpenAI shadow
The backdrop is awkward. Seven people who left OpenAI founded Anthropic in 2021, reportedly doubting that the older lab could build powerful models safely. This September, OpenAI, which had quietly filed confidentially over the summer, said it was dropping plans to go public this year, citing the dangers AI poses to humanity.
So the rival the founders walked away from is sitting out the market while the safety-first spinoff walks in. CEO Dario Amodei has repeatedly cautioned that AI could eliminate vast numbers of jobs and work against human interests. Now he has to sell shares on that candor.
The filing ends its pitch by urging buyers to look past the deal itself toward "something greater." Investors have heard that before, usually just ahead of a lockup expiring.
Watch for three things. First, an amended S-1 with a price range and share count, which will show whether the $2 trillion talk holds up. Second, how the supply-chain lawsuits move before the listing date. Third, whether OpenAI's decision to wait starts to look prudent or timid once ANT actually trades.