Anthropic IPO Filing Shows $42B Loss, $518B AI Bet

Anthropic's confidential IPO paperwork became a little less confidential this week. A prospectus reviewed by Reuters shows the AI lab lost $42 billion in 2025 on just $4.6 billion of revenue, and has locked in $518 billion worth of cloud and chip commitments for the year ahead. Anthropic is framing itself, in its own filing, as a company betting AI reshapes the economy more than industrialization or the internet did. The numbers say that bet is currently very expensive.

• Net loss of $42 billion in 2025, roughly nine times the company's total revenue for the year.
• Revenue grew 12-fold to about $4.6 billion, while operating losses topped $8 billion even before writedowns tied to earlier fundraising.
• Compute and infrastructure spending tripled to $7.33 billion, more than half of $12.65 billion in total operating expenses.
• Anthropic has committed to $518 billion in cloud, compute and infrastructure obligations going forward, and its public debut looks likely to slip past November's US midterms.

What exactly is in the prospectus?

The document lays out a company growing fast and burning cash faster. Revenue for 2025 came in near $4.6 billion, a twelvefold jump from the year before, driven by Claude's enterprise coding and agent products. Total operating expenses hit $12.65 billion, and compute alone (the GPUs, TPUs and cloud contracts needed to train and serve Claude) ate $7.33 billion of that, up roughly 3x year over year. Strip out one-time writedowns tied to earlier funding rounds and Anthropic still lost more than $8 billion on an operating basis. Add those writedowns back in, mostly fair-value changes on convertible instruments and warrants from prior raises, and the net loss balloons to $42 billion.

RelatedNscale Files for NYSE IPO, Sitting on $103.4B in Contracts

The headline number that will follow Anthropic into its roadshow is the $518 billion figure: cloud, compute and infrastructure obligations the company has already committed to, spanning multiple years of contracts with its cloud partners. That is not cash spent in 2025. It is a forward liability, the kind of multi-year capacity reservation that has become standard practice for frontier labs racing to lock in scarce GPU and TPU supply before a competitor does.

Why go public with a $42 billion loss?

Because the loss is a side effect of the bet, not evidence against it. Anthropic's prospectus argues AI will remake the economy more thoroughly than electrification or the internet did. Whether that framing holds up, the growth is real: revenue up twelvefold in a year, with an annualized run rate reportedly past $40 billion by mid-2026, up from roughly $9 billion at the end of 2025. Investors here aren't underwriting 2025's numbers, they're underwriting that curve continuing, the same bet that took Amazon and Uber public while both were deeply unprofitable.

The ask has grown with the story. Anthropic's Series H in mid-2026 priced the company at $965 billion post-money. The prospectus reportedly targets more than double that, putting Anthropic in range of a $2 trillion listing before it has ever filed a profitable quarter.

Can a company actually spend $518 billion and survive?

Not out of its own pocket, and it isn't trying to. That figure is a multi-year commitment spread across cloud partners (Amazon and Google are both investors and infrastructure providers, alongside custom silicon deals), not a single check Anthropic has to write this year. The mechanism is the same one every frontier lab now uses: sign long-dated capacity contracts now, at today's negotiated rates, to guarantee GPU and TPU supply before a competitor locks it up first. It's a bet that demand for Claude keeps climbing fast enough to make good on the contract, financed increasingly through debt and specialized infrastructure vehicles rather than pure equity, which is exactly why the IPO matters: public markets are the next, much larger pool of capital these commitments assume will be there.

What does the prospectus say about the IPO timeline?

It's slipping. As recently as September, reporting pegged Anthropic's target as pricing before November's US midterm elections, with Nvidia in talks to put up to $10 billion in as an anchor investor. The prospectus now reviewed by Reuters points to a debut likely pushed past the midterms instead, which tracks with a filing this large and financially messy needing more time with underwriters and regulators, not less.

RelatedNvidia May Put $10B Into Anthropic's Record IPO

1. Jun 1, 2026Confidential draft S-1 filed with the SEC following the $965B Series H
2. Sep 12, 2026Nvidia anchor talks reported up to $10B, target pricing before the midterms
3. Sep 28, 2026Prospectus details leak to Reuters $42B net loss, $518B in commitments disclosed
4. Late 2026 / early 2027Public debut expected now likely after November's midterms

What it means for the market

The most useful comparison isn't hypothetical, it's OpenAI, reportedly heading toward its own listing on a nearly identical shape: about $13.1 billion in 2025 revenue against a $41.55 billion net loss, driven by similar fair-value writedowns tied to its for-profit restructuring. Both labs are entering public markets with losses that roughly match, or exceed, all-time revenue, asking investors to price them on 2027-2028 growth instead of trailing financials. For Nvidia, Amazon, Google and Microsoft, simultaneously investors, infrastructure suppliers and (via anchor-investment talk) prospective IPO backers, the read is straightforward: their upside depends on Anthropic converting that $518 billion commitment into paying customers, not just training runs. The signal for investors is less "is Anthropic profitable" and more "does the compute commitment scale with real revenue, or is this the AI-infrastructure equivalent of a lease nobody can walk away from."

Who does this actually affect?

Anthropic employees holding equity are the most directly exposed: a $2 trillion listing versus a $965 billion private mark roughly doubles paper wealth overnight, assuming the market accepts the price. Cloud and chip suppliers (Amazon, Google, and whichever GPU vendors are inside that $518 billion figure) have a direct interest in the listing succeeding, since a failed or down-priced IPO would cast doubt on whether those multi-year contracts get paid down as scheduled. Rival labs, especially OpenAI and its own IPO ambitions, will be watched against Anthropic's pricing as the market's first real referendum on whether frontier-AI losses at this scale are investable at all. And everyday Claude users and enterprise customers are unlikely to see any near-term change: the spending disclosed here funds capacity Anthropic already believes it needs, not a shift in product strategy.

Our take

A $42 billion loss sounds catastrophic until you notice OpenAI just posted almost the same number on a comparable revenue base, and both companies are still finding buyers willing to underwrite trillion-dollar-plus valuations anyway. That doesn't make the loss meaningless. It means the market has, for now, decided to price frontier AI labs on growth rate and compute lock-in rather than trailing profitability, the same way it once priced cloud infrastructure buildouts. The $518 billion commitment is the number worth watching more than the $42 billion loss: a loss is a snapshot, but a multi-year infrastructure obligation is a bet the company can't easily walk back if Claude's growth curve bends before the contracts do. Anthropic's prospectus reads like a company confident that bet pays off. Whether public investors agree at a $2 trillion price is the actual test, and it's a test that won't be graded until well after these midterms.

Original analysis by GenZTech. Source: Reuters.