Anthropic's Real Risk Isn't The $42 Billion Loss

Anthropic's secret IPO prospectus leaked this week, and the headline number, a $42 billion net loss in 2025, is designed to grab attention. According to new Reuters reporting broken down on this episode of Good Revenue, that number is mostly a magic trick. But the real figures buried underneath tell a much more interesting story about what Dario Amodei's company is actually betting on.

The Loss Isn't What It Looks Like

Roughly 80 percent of that $42 billion, about $34 billion, is an accounting charge reflecting the rising estimated value of financing that could eventually convert into Anthropic shares. It's a paper adjustment, not cash that left the building. That should calm nerves. It shouldn't calm them for long.

Strip away the accounting noise and Anthropic still lost more than $8 billion on an operating basis in 2025, up from $2.98 billion the year before. That's a threefold increase in real cash losses in a single year. This is the number that matters, and it's heading in the wrong direction for a company trying to convince public market investors it deserves a $2 trillion valuation.

Where The Money Is Actually Going

Compute is the answer. Anthropic spent $7.33 billion on compute in 2025, triple what it spent the year before, and more than half of its total $12.65 billion in operating expenses. This is coming from a company that has publicly positioned itself as the voice of caution in the AI arms race, arguing for slower, more responsible spending. The prospectus tells a different story.

The real shocker is what's coming next. Anthropic has committed to $518 billion in compute, cloud, and infrastructure costs in the years ahead. That's twelve times the size of the loss everyone is going to be talking about. It's the number that should actually shape how investors think about this IPO, because it signals a company with no near-term plan to bend its cost curve downward.

Growth Is Real, But So Is The Math Problem

To be fair, Anthropic's revenue told a genuinely good story: $4.6 billion in 2025, up from $400 million in 2024, a twelvefold jump. But that revenue trajectory is being asked to justify a valuation of $2 trillion, more than double where the company stood just months ago in May. The math is a stretch, and the market has reason to be skeptical. SpaceX's IPO debut turned into a rapid unwind within a month, and OpenAI, Anthropic's chief rival, appears to be struggling too, with a underwhelming model launch, paused training runs, and reports of a scramble for fresh private funding.

Anthropic, notably, has been more stable than OpenAI despite being roughly half its age. That stability is real. But the prospectus also discloses something harder to quantify: executives and outside analysts have raised serious concerns about AI's potential negative consequences, even existential ones. That a company has to formally disclose this risk in its own IPO filing is a moment worth sitting with.

What Comes Next

Reuters reports Anthropic is targeting a mid-November IPO, timed just after the U.S. elections. The real ask embedded in that filing isn't for investors to look past a big loss. It's for them to commit to a massive, open-ended infrastructure bet on compute, one with no clear end date and no evidence yet that the spending will slow. Whether investors are willing to make that leap in November will say a lot about where the entire AI sector goes from here.