Anthropic IPO Filing Shows $42 Billion Loss, $4.6B Revenue

Anthropic's confidential draft IPO prospectus shows 2025 revenue surged roughly twelvefold to nearly $4.6 billion, while the company posted a net loss of about $42 billion — but roughly $34 billion of that loss was a non-cash accounting charge, not cash burned running the business, according to Reuters' review of the filing. Strip out that charge and Anthropic's actual operating loss was about $8 billion, still more than double 2024's, as the AI lab piles up infrastructure commitments reported at $518 billion.

What Reuters' Review of the Prospectus Found

Anthropic confidentially submitted a draft registration statement on Form S-1 to the Securities and Exchange Commission on June 1, 2026, a step that gives the AI company the option to go public once SEC review wraps up but commits it to nothing. Anthropic's own announcement of the filing said the number of shares and the price had not been set, and that any offering would depend on market conditions. For months afterward, the document's contents stayed private, as JOBS Act-style confidential submissions are designed to.

That changed in late September 2026, when Reuters reported it had reviewed the draft prospectus and found a far more detailed financial picture than Anthropic had disclosed publicly up to that point. Pandromeda previously covered the broad strokes of Anthropic's IPO timeline, valuation chatter and filing status; this is a closer look at what the financial disclosures Reuters reviewed actually say about the business underneath the headline numbers — revenue, losses, spending commitments, customer risk and cash on hand.

Revenue Grew Twelvefold, to Nearly $4.6 Billion

According to Reuters, Anthropic's 2025 revenue came in at just under $4.6 billion, up roughly twelvefold from about $386 million in 2024. That growth rate reflects the compounding effect of enterprise and developer adoption of Anthropic's Claude model family, including large API and cloud-platform deals, through a year in which Anthropic also pushed deeper into tools aimed at professional and engineering users. The company's wider push to court enterprise technical talent — including the $100 million training initiative detailed in Pandromeda's look at the Claude Frontier Academy program — sits alongside that commercial expansion as part of Anthropic's broader bet on enterprise and developer demand.

Twelvefold growth is an extraordinary rate for a company already generating billions in revenue, and it's the number Anthropic and its bankers are likely to lean on hardest in any eventual roadshow. But the prospectus, as Reuters describes it, pairs that growth with an even faster rise in costs — which is where the loss figures come in.

Breaking Down the $42 Billion Net Loss

The headline figure in Reuters' reporting is a 2025 net loss of roughly $42 billion — specifically about $41.97 billion. Taken at face value, that would make it one of the largest annual losses ever disclosed by a private company ahead of an IPO. But Reuters' review makes clear that the number is misleading if read as straightforward cash burn.

About $34 billion of that net loss, per Reuters, was a non-cash accounting charge tied to the rising estimated value of financing instruments — stock the company had previously issued that could convert into Anthropic shares. As those instruments' value climbed alongside Anthropic's soaring valuation (from roughly $965 billion in May's funding round toward a reported IPO target north of $2 trillion), accounting rules required Anthropic to book the increase as an expense, even though no cash left the company. It's a mark-to-market-style charge rather than a reflection of operating performance, and it's a dynamic common among richly valued private companies that raised capital through convertible instruments.

This kind of charge tends to show up precisely when a company's valuation rises fastest, which can make the mechanics feel counterintuitive: the better Anthropic's fundraising went, the larger the paper loss on its income statement. Investors evaluating the eventual public filing will likely look past the net-loss headline to the operating figures instead, since that's the number that reflects how the underlying business is actually performing — and it's also the number that public-market analysts will model forward when sizing up a path to profitability.

It's that distinction — a multibillion-dollar accounting artifact sitting on top of a separate, genuine operating loss — that makes this filing's financial disclosures more nuanced than the eye-catching $42 billion figure alone suggests.

Strip that charge out, and the picture looks very different — though still not profitable.

The Real Number: An $8 Billion Operating Loss

Excluding the non-cash charge, Reuters reported Anthropic's actual operating loss for 2025 at about $8.06 billion, up from roughly $2.98 billion in 2024. That's still a loss that more than doubled year over year, even as revenue grew far faster. The gap is largely a function of compute: Reuters' review put Anthropic's 2025 compute and infrastructure spending at $7.33 billion, roughly three times what it spent in 2024, and equal to about 58% of the company's $12.65 billion in total operating expenses for the year.

In other words: Anthropic is still losing money even before the accounting noise, and it's losing more of it each year, because running and scaling frontier AI models remains extraordinarily compute-intensive. That's consistent with what rival labs have said about their own economics, and it underscores why Anthropic's spending commitments — the next item in the prospectus — matter so much to how investors will size up the IPO.

A Reported $518 Billion Spending Commitment

Beyond what it already spent in 2025, Reuters reported that Anthropic's prospectus discloses roughly $518 billion in planned or committed cloud, compute and infrastructure spending in the years ahead — contractual obligations the company has locked in with cloud and chip partners to secure the capacity it believes it will need to keep training and serving frontier models. That figure dwarfs Anthropic's current cash position and even its reported revenue run-rate, and it illustrates just how capital-intensive the frontier AI race has become for every major lab, not just Anthropic.

It's also a reminder that Anthropic's spending commitments extend well beyond its own data centers and models. The company has separately pledged money toward initiatives like the government-facing $150 million Genesis Mission science push Pandromeda covered in October, on top of the compute and cloud obligations detailed in the prospectus — all part of a broader spending posture that any IPO prospectus will need to justify to public-market investors accustomed to clearer paths to profitability.

Customer Concentration: A Quarter of Revenue From Two Clients

One of the more pointed risk disclosures Reuters found in the draft prospectus: nearly a quarter of Anthropic's 2025 revenue came from just two customers. The filing reportedly also warns that many of Anthropic's largest enterprise customers are not locked into long-term contracts, meaning they could reduce or halt spending with comparatively little friction.

That combination — heavy reliance on a handful of large accounts, paired with limited contractual lock-in — is precisely the kind of risk factor that public-market investors and underwriters scrutinize closely in any tech IPO, and it will likely draw particular attention given how much of Anthropic's revenue growth has come from enterprise and cloud-platform deals rather than a broad base of smaller customers. It's also a reminder that strong headline revenue growth doesn't by itself tell you how durable that revenue is.

Concentration risk of this kind is common among fast-scaling AI infrastructure providers — a small number of hyperscalers and large enterprises often account for an outsized share of early revenue simply because they're the customers with the budgets and workloads to absorb frontier-model pricing at scale. But it still matters for how a prospectus gets priced: a company whose growth leans on a few large, cancellable relationships typically draws a more conservative valuation multiple than one with broad-based, contractually locked-in demand, all else being equal. Expect this risk factor, once the filing is public, to be one of the more closely parsed sections by equity analysts weighing Anthropic's reported $2 trillion-plus ask.

Cash Position: $20.28 Billion on Hand

On the balance-sheet side, Reuters reported that Anthropic held $20.28 billion in cash, cash equivalents and short-term investments as of December 31, 2025. That gives the company a meaningful cushion relative to its reported operating losses, but it's a small fraction of the $518 billion in infrastructure obligations reported elsewhere in the filing — underscoring that Anthropic is relying on a mix of future revenue growth, additional fundraising, and the IPO itself to fund its spending plans, rather than sitting on enough cash today to cover them outright.

Valuation and Timing: Reported Plans, Not Confirmed Facts

It's worth being precise about what is and isn't confirmed here. Reuters and other outlets have reported that Anthropic is seeking a valuation above $2 trillion — more than double the roughly $965 billion valuation it reached in its May 2026 funding round — and that any public listing is more likely to come after the November 2026 US midterm elections than before. Anthropic declined to comment to Reuters on the prospectus's contents, including the loss figures, spending commitments and valuation target.

Those valuation and timing details should be read as reported plans sourced to people familiar with the process, not as confirmed facts or anything Anthropic has stated on the record. Pricing, timing and even the decision to proceed with a public offering at all can change between a confidential draft filing and an actual listing — sometimes substantially.

Why the Prospectus Still Isn't on SEC EDGAR

Because Anthropic used the confidential, JOBS Act-style draft-registration process available to emerging growth companies, its draft S-1 is not publicly viewable. A search of the SEC's EDGAR system for Anthropic turns up no public S-1 or related registration filings. That's expected and by design: confidential draft submissions let a company get SEC staff feedback on a registration statement before any of its contents — including the financial disclosures Reuters reviewed — become public. The document typically becomes public only shortly before a roadshow begins, if the company proceeds to list at all. For background on Anthropic as a company, including its founding, funding history and governance structure, see Wikipedia's overview of Anthropic.

What to Watch Next

The next concrete milestone will be whether Anthropic's draft S-1 is amended and then filed publicly — the point at which its financials, risk factors and governance structure (reportedly including a founder-controlled voting entity) would become verifiable primary-source documents rather than reported figures. Separately, Bloomberg Government has reported that Anthropic's annualized revenue run-rate climbed from about $9 billion at the end of 2025 toward $65 billion by late July 2026, a different and faster-moving metric than the audited annual figures in the prospectus, and one Anthropic has not confirmed in detail. Expect scrutiny to center on three things once more becomes public: whether the operating-loss trajectory bends toward breakeven as Anthropic has reportedly told some investors it expects in 2026, whether the customer-concentration risk narrows as adoption broadens, and whether the eventual IPO price reflects the more-than-$2 trillion valuation that has so far only been reported, not confirmed.

Until Anthropic files publicly, the numbers in this piece should be treated the way they're sourced: as Reuters' account of a confidential draft document, not as audited figures Anthropic has put its own name to.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage