Anthropic's IPO arrives as the AI market slides: investors will price the gap between revenue and losses

By Benzinga · October 31, 2026.

Anthropic is preparing an initial public offering (IPO) that could come as early as November, according to Benzinga. The backdrop is far from ideal: many AI stocks have fallen sharply, the outlet says.

The most telling figure comes from the company's prospectus, cited by Benzinga. In the first half of the year, Anthropic posted revenue of $16.8bn and losses of $80bn. The outlet adds that the company expects to keep spending heavily in the coming years. Its losses for the period were almost five times its revenue.

Benzinga also points to other signs of cooling:

- Z.ai and MiniMax, listed in Hong Kong and comparable because they build models, have fallen more than 80% from their highs. - SpaceX, which went public in June, is trading 40% below its peak. The outlet ties it to AI through Grok and its data center rental business. - Firmus, a data center company backed by Nvidia, cancelled its IPO after investors rejected a $30bn valuation. Nscale, also backed by Nvidia, is considering cutting its $33bn valuation. - OpenAI, according to Benzinga, raised $11bn of the $30bn it was seeking. - Blackstone has warned that Anthropic could struggle to secure an investment-grade credit rating after the IPO, according to the outlet. That would keep its borrowing costs higher over the long term.

The article adds that, according to Polymarket, most bettors expect Anthropic to be valued at $1.5 trillion in December. This is a market prediction, not a company figure.

Our take: the underlying thesis is that this IPO will not be a referendum on whether Claude is any good, but on how long a capital race can keep being financed. Two things need to be kept apart. One is the quality of the product. The other is the company's economics, where spending several times what comes in only works as long as the market is still willing to pay for the future.

One important caveat: Benzinga is a markets outlet, and the headline plays up the risk. A slump in comparable companies does not prove the IPO will go badly. And Blackstone's warning about the rating is the opinion of a financial player, not a done deal. Also, the half-year losses of a company investing in compute are not the same as operating losses: they depend on how much of the spending goes to training and infrastructure. The prospectus is where to check that.

In general, tech bubbles leave behind useful infrastructure even when they wipe out valuations. It happened with the railways and with fiber optics in the early 2000s. In the short term, a correction would hurt: less cheap capital, layoffs and pressure to monetize sooner, with effects on jobs across the sector. But in the long term, the compute capacity and the models already built do not disappear. What is at stake is the pace, not the direction.

Who wins and who loses if the market tightens: the losers are labs that rely on ever-larger funding rounds. The winners are companies with solid recurring revenue and access to cheap capital, and also open-weight models, whose cost of use is falling just as closed-model pricing faces more scrutiny. That is why, rather than the share price on day one, we will be watching three things: gross margin per token, the trend in recurring revenue, and the credit rating the company ends up with.