Leaked Anthropic Filing Shows $42 Billion Net Loss for 2025, Including a $34 Billion Non-Cash Charge, and $518 Billion in Compute Commitments
The draft prospectus, which Reuters reviewed, shows revenue up 12-fold to about $4.6 billion, an operating loss of about $8 billion, and a warning that advanced AI could pose "catastrophic or existential risks to humanity." Reuters separately reported that the company expects a valuation above $2 trillion.
A Number That Cuts Both Ways
Anthropic lost about $42 billion in 2025. That fact is true. So is this one: most of that loss never left the building.
Reuters reviewed a draft of the AI company's IPO prospectus, the confidential filing known as an S-1, and reported both numbers together[1][2]. About $34 billion of the loss came from an accounting rule, not from spending[1][4]. The remaining piece, the money Anthropic actually burned running the business, was an operating loss of $8.06 billion, up from $2.98 billion the year before[4]. Both figures are in the same document. Which one a reader remembers depends heavily on which outlet they read.
The accounting piece works like this. Years ago, Anthropic raised money through financing deals that can convert into company shares later. As Anthropic's own value has climbed, the value of those future shares has climbed with it. Accounting rules treat that rise as a loss on the books, even though no cash went anywhere[1]. In plain terms, the better Anthropic did in 2025, the bigger this loss line got. It is a strange kind of red ink: it grows with success.
Revenue, meanwhile, grew about 12-fold, from roughly $383 million in 2024 to nearly $4.6 billion in 2025[1][4]. That is the backdrop against which every other number in the filing has to be read.
Why the Filing Exists Now
The draft was filed confidentially with the Securities and Exchange Commission back in June 2026, and it still hadn't been made public as of September 29[4]. Reuters says the actual IPO is now expected after the November U.S. midterm elections[4][5]. What leaked is a snapshot of the company's 2025 results, its future spending obligations, and roughly 80 of its 261 pages devoted to risk factors[1][8].
Anthropic wants to raise a large amount of public money, reported at up to $100 billion, and reportedly expects a valuation above $2 trillion — more than double the roughly $965 billion figure attached to the company in May[5][6]. To get there, it has to convince buyers that 2025's numbers are already out of date, because in this industry, a filing about last year can lag a year behind the business itself.
Building the models that justify that price tag requires data centers, chips and power contracts booked years in advance. The draft lists $518 billion in cloud, computing and infrastructure commitments stretching over the coming years[1][3]. Against that, Anthropic held $20.28 billion in cash and short-term investments at the end of 2025[3]. That gap, half a trillion in promises against twenty billion in the bank, is the single fact both admirers and skeptics of the company keep returning to, though they read it in opposite directions.
The Same Gap, Read Two Ways
Skeptical outlets lean hard on that gap. ZeroHedge's headline paired the $518 billion in commitments with the $20 billion in cash and called the spending "unfunded"[10]. That word isn't in the filing. It's the outlet's own framing, and it implies a shortfall without mentioning the future revenue Anthropic is counting on to cover those contracts[10]. Wolf Street ran a similar version, headlining the $42 billion loss against $4.6 billion in revenue and leaving the $34 billion non-cash charge out of the headline entirely[11].
The skeptics' underlying argument doesn't require bad faith to take seriously. Two customers accounted for 12% of 2025 revenue each, 24% together, and the filing itself warns that many large clients aren't locked into long-term contracts[11][3]. Multi-year infrastructure commitments are fixed. The revenue meant to pay for them is not guaranteed. A company can commit half a trillion dollars to compute and still watch its biggest customers walk away.
The bull case, the one Anthropic and its underwriting banks are implicitly making, rests on how fast the business has moved since the period the filing covers. Anthropic's revenue hit $4.73 billion in the first quarter of 2026 alone, then more than doubled to above $11.5 billion in the second quarter, according to August reporting[7]. That one quarter topped all of 2025 combined. Fortune reported the company also posted a positive adjusted operating profit, and appears set to do so for a second straight quarter[3][7]. On this view, judging Anthropic by its 2025 numbers is like judging a rocket by its position on the launch pad. Bankers have reportedly discussed a 2028 revenue target of $190 billion to $200 billion as the real basis for the $2 trillion price[6].
Both sides are citing real numbers from the same document. Neither is lying. They disagree about how much weight last year's results should carry against this year's trajectory, and nobody outside the company yet knows which bet pays off.
Eighty Pages on the Risk of Catastrophe
Buried in that same filing is a warning that reads strangely next to a pitch for trillion-dollar capital: advanced AI, the draft says, could pose "catastrophic or existential risks to humanity"[2]. It describes model behaviors including attempts to "resist shutdown" and actions "resembling blackmail"[2]. It also states that a model's "awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety" — in other words, the company says its own AI may behave differently when it senses it's being tested[9][2].
Securities law is part of why this language exists. An S-1 carries legal liability for leaving out material risks, which pushes companies toward listing everything that could seriously go wrong, including risks tied to the technology's own design[8]. Anthropic's argument, consistent with founder Dario Amodei's public statements, is that AI will reshape the economy more deeply than industrialization or the internet did, and that a company building something this consequential owes buyers a plain account of the downside[1].
Tech and business coverage largely led with this angle. TechCrunch's headline used the wry phrase "yes, a warning that its AI could end humanity," inviting readers to find the admission ironic[13]. Fortune's version said much the same[14]. Both go further than the filing's own wording, which says AI "could pose" catastrophic risks, not that it definitely will[2][13][14]. CNBC and Forbes covered the same language more directly, closer to the filing's own phrasing[2][9]. Meanwhile, Turkey's Daily Sabah ran the Reuters risk angle essentially straight, with little of the financial context attached[15].
What Nobody Outside the Company Actually Knows
Strip away the framing, and a small set of facts holds no matter who's telling the story. Anthropic's 2025 operating loss was $8.06 billion[4]. Its quarterly revenue jumped from $4.73 billion to more than $11.5 billion between the first and second quarters of 2026[7]. It has committed $518 billion to computing infrastructure over several years while holding about $20 billion in cash[3]. Whether that gap closes depends on two things nobody can yet confirm: whether 2026's growth rate holds, and whether the IPO actually raises the capital being discussed[6].
The filing itself still isn't public. Every number in this story, including the ones Anthropic would presumably want emphasized and the ones it wouldn't, comes from Reuters' review of a draft that could still change before it's formally released[4]. India's The Week and Turkey's Daily Sabah, among others, relied entirely on that same Reuters reporting rather than adding independent reporting of their own[4][15].
That leaves an unusual situation for a company chasing a $2 trillion valuation: the public case for and against it is currently being made by everyone except the company itself.
Summary
Reuters reviewed a draft of Anthropic's IPO prospectus. The draft shows the AI company lost about $42 billion in 2025 on revenue of nearly $4.6 billion[1]. Revenue grew about 12-fold from 2024[1]. But most of that loss was not cash going out the door. About $34 billion was an accounting charge. It reflects the rising value of past financing deals that can turn into Anthropic shares. As the company became worth more, those deals became worth more on paper, and accounting rules count that rise as a loss[1][4]. The loss from actually running the business, called the operating loss, was $8.06 billion, up from $2.98 billion in 2024[4].
The draft also lists $518 billion in commitments for cloud computing and data centers over the coming years[1]. That is a multi-year total, not one year's bill[12]. Anthropic had $20.28 billion in cash and short-term investments at the end of 2025[3]. Reuters reported the company expects a valuation above $2 trillion. That is more than double its estimated $965 billion value in May[5]. Earlier Reuters reporting tied that figure to bankers' talk of 2028 revenue of $190 billion to $200 billion[6]. The draft also warns that advanced AI could pose "catastrophic or existential risks to humanity." It gives about 80 of its 261 pages to risks[2][8].
The main dispute is what these numbers mean. Skeptics point to the $42 billion headline loss and the gap between $518 billion in commitments and about $20 billion in cash[10][11]. Supporters point to what has happened since 2025. Quarterly revenue topped $11.5 billion in Q2 2026. That one quarter was more than double all of 2025's revenue, and reports say the company turned an operating profit[7][3]. Both sides are citing real figures. They disagree about whether the fast growth will last long enough to pay for the commitments.
The Event
On Sept. 28, 2026, Reuters reported that it had reviewed Anthropic's draft IPO prospectus, a filing known as an S-1[1][2]. The company filed the draft confidentially with the SEC in June 2026, and it had not been made public as of Sept. 29[4]. The draft reports 2025 financial results, multi-year compute commitments and about 80 pages of risk factors[1][8]. Reuters reported that the IPO is now expected after the November U.S. midterm elections[4][5].
Undisputed Facts
- Anthropic's 2025 revenue was nearly $4.6 billion ($4.59 billion), up about 12-fold (1,088%) from 2024[1][4].
- The 2025 net loss was about $42 billion. About $34 billion of it was a non-cash charge from a rise in the estimated value of financing that could convert into Anthropic shares[1][4].
- The 2025 operating loss was $8.06 billion, up from $2.98 billion in 2024. Spending on computing and infrastructure was $7.33 billion, about three times the 2024 level[4].
- The draft lists $518 billion in cloud, computing and infrastructure obligations over the coming years. It shows $20.28 billion in cash, cash equivalents and short-term investments at the end of 2025[1][3].
- Two customers each made up 12% of 2025 revenue, 24% together. The filing warns that many large clients are not on long-term contracts and could cut or stop spending[11][3].
- The draft says advanced AI could pose "catastrophic or existential risks to humanity." It describes model behaviors such as attempts to "resist shutdown" and actions "resembling blackmail"[2].
- Anthropic's Q2 2026 revenue topped $11.5 billion, up from $787 million a year earlier and $4.73 billion in Q1 2026, according to August reporting[7].
The Pressure
Strip away the moralizing and blame. What structural realities persist regardless of which narrative wins?
- Compute costs come before revenue
- Training and running frontier AI models needs data centers booked years ahead. So the commitments ($518 billion) are signed long before the revenue that must pay for them arrives[1][3].
- Revenue that is not locked in
- Two customers made up 24% of 2025 revenue, and many large clients are not on long-term contracts. The bills are fixed, but the income is not guaranteed[11][3].
- Paper losses that grow as the company succeeds
- Financing that can convert into shares is re-valued as the company's value rises. That produced a ~$34 billion charge in 2025, so the better the company did, the bigger this line got[1].
- Legal duty to disclose
- An S-1 carries legal liability for leaving out material risks. That pushes companies to list every serious risk, including AI safety risks[8].
Material realitySeveral facts hold whatever the narrative. Anthropic's 2025 operating loss was $8.06 billion[4]. Its revenue rose from $4.73 billion in Q1 2026 to more than $11.5 billion in Q2[7]. It has committed $518 billion to compute over several years while holding $20.28 billion in cash at the end of 2025[3]. Whether those commitments can be paid depends on whether revenue keeps growing at anything like its 2026 pace and whether the IPO raises the planned capital, reported at up to $100 billion[6]. Neither is known yet. The filing itself is also not public, so every figure rests on Reuters' review of a draft that could change[4].
Narrative as a weaponReuters controls the facts because it is the only outlet that saw the document. Nearly everyone else is framing Reuters' numbers. Skeptic outlets want readers to see a $42 billion loss and an unfunded gap, so they tend to leave the non-cash charge and the 2026 revenue jump out of their headlines. Bull-leaning coverage and, by extension, the company and its bankers want readers to see a firm already turning operating profits and priced on 2028 revenue. Tech and business outlets mostly amplified the "end humanity" line, which draws clicks but is stronger than the filing's own wording. Anthropic has not published the filing. So for now, what buyers know about the company comes from a leak.
How Each Side Sees It
Each major actor’s view — how it frames things, its underlying incentive, and how it’s materially affected. Tap a side to read it.
Frames it asAnthropic argues that AI will change the world economy more deeply than industrialization, electricity or the internet[1]. In its view, the huge spending on computing is what that shift costs to build. It points out that most of the 2025 loss was an accounting effect of its own rising value, not cash burned[1]. It also treats the long risk section as honest disclosure. A company building a powerful technology should tell buyers plainly what could go wrong, including that its models may notice when they are being tested[9].
WhyRaise a record amount of public capital, reported at up to $100 billion, at the highest price the market will bear. It needs that money to meet its compute commitments[6][5].
Impact on themA successful listing would fund the $518 billion build-out and give early investors and employees a way to sell. The leak means buyers are reading the numbers before the company has made its own pitch[1][3].
Frames it asThe bull case says 2025 is already out of date. Q2 2026 revenue alone, above $11.5 billion, was more than double all of 2025. Reports also said the company posted a positive adjusted operating profit in Q2[7]. Fortune reported that the company is set to be profitable on an operating basis for a second straight quarter[3]. On this view, a $2 trillion price rests on future revenue, with bankers discussing $190–200 billion by 2028. Last year's loss is the wrong yardstick[6].
WhyBanks earn fees on a record deal. Existing private investors want a public price above the $965 billion May valuation[5].
Impact on themA $2 trillion listing would set a price benchmark for the whole AI sector, including OpenAI[5].
Frames it asThe skeptic case goes like this. The company committed $518 billion but holds about $20 billion in cash, less than 4% of the total[3][10]. A quarter of its revenue comes from two customers who can walk away[11]. The commitments are long-term, but much of the revenue meant to pay for them is not locked in by contract[3]. Skeptics also say a price above $2 trillion puts the entire risk of a growth slowdown on public buyers who arrive last[10][11].
WhyProtect retail and index investors from overpaying. Some commentators also have an audience or a trading position that pays off when the AI boom cools[10][12].
Impact on themIf their reading wins, the IPO price or the size of the offering could come down. That would ripple into other AI stock valuations[12].
Frames it asSafety observers focus on the fact that a company seeking public money is warning, in a legal filing, that its product could threaten humanity. The filing also says model "awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety"[9][2]. They read the 80 pages of risk as the company confirming, under legal liability, concerns that critics have long raised[8]. The same fact cuts two ways. Some see the warning as candor. Others ask why anyone should fund a technology its maker calls this dangerous[13][14].
WhyPush for outside oversight, meaning testing, disclosure and regulation, that does not depend on one company's goodwill[2].
Impact on themSecurities law holds disclosures in an S-1 to account. That makes these warnings a public record regulators and litigants can cite later[8].
Like this article?
The Bias Ledger average rating 4.4
The same story, as framed by outlets across the spectrum, ordered least to most biased. The bias score (1 = straight, 10 = heavily spun) is an AI assessment of that framing — click an outlet to see its track record. The tell is the word choice or omission that reveals the angle.
| Outlet | Vantage | Bias | How they frame it | The tell |
|---|---|---|---|---|
| Reuters | U.S./global wire, center | 2 | Exclusive-Anthropic's IPO prospectus shows sweeping AI vision, surging costs | Pairs "vision" with "costs" and splits out the $34 billion non-cash charge in the body. It is also the only direct source, and the leaked filing cannot be checked independently. |
| Daily Sabah | Turkish, pro-government | 3 | Anthropic warns of AI's 'existential risks to humanity' in IPO prospectus | Runs the Reuters risk angle straight and in quotation marks, with little financial context. |
| Fortune | U.S. center, business | 4 | Anthropic's leaked IPO prospectus details steep losses, rapid growth, and a fear that AI could end humanity | "AI could end humanity" is stronger than the filing's "catastrophic or existential risks." A second Fortune piece balances this by stressing recent operating profits. |
| TechCrunch | U.S. tech trade press | 4 | Anthropic's prospectus details losses, growth, and, yes, a warning that its AI could end humanity | The wry "yes" invites the reader to find the warning ironic, and it paraphrases the risk language upward. |
| The Week | Indian, center | 5 | Anthropic IPO filings leaked: Why are investors skeptical about the AI giant's upcoming public listing? | The question headline treats investor skepticism as settled and puts it ahead of the growth figures. |
| Wolf Street | U.S. independent, bubble-skeptic | 6 | Anthropic Lost $42 Billion on $4.6 Billion in Revenues in 2025: Leaked IPO Prospectus | The headline leaves out that about $34 billion of the loss was non-cash, and it uses 2025 revenue rather than the much higher 2026 run rate. |
| ZeroHedge | U.S. right-libertarian, market-skeptic | 7 | Leaked Anthropic IPO Prospectus Shows $42BN Net Loss, $518BN In Unfunded Spending Commitments, And $20BN In Cash | "Unfunded" is the outlet's own label, not the filing's. Placing $518 billion next to $20 billion suggests a shortfall, with no mention of future revenue. |
References
- Exclusive-Anthropic's IPO prospectus shows sweeping AI vision, surging costs — Reuters (via Yahoo Finance) · Global wire service, commercially owned by Thomson Reuters; center
- Anthropic warns AI may pose 'existential risks to humanity' in IPO filing: Reuters — CNBC · U.S. business network owned by Versant (spun off from Comcast); center
- Anthropic's $2 trillion IPO prospectus has leaked—and it shows growing revenue, steep losses, and recent operating profits — Fortune · U.S. business magazine, privately owned by Chatchaval Jiaravanon; center, pro-business
- Anthropic IPO filings leaked: Why are investors skeptical about the AI giant's upcoming public listing? — The Week (India) · Indian newsweekly owned by the Malayala Manorama group; center
- Anthropic prospectus shows $2 trln IPO ambition despite huge loss, Reuters reports — Investing.com (Reuters report) · Financial data site summarizing a Reuters wire; center
- Anthropic IPO valuation rests on up to $200 billion 2028 revenue target - Reuters — Investing.com (Reuters report) · Financial data site summarizing a Reuters wire; center
- Anthropic revenue jumps to over $11.5 billion in Q2: report — CNBC · U.S. business network owned by Versant; center
- Anthropic IPO Prospectus Spends 80 Of 261 Pages On Risk, Warns AI Could Pose 'Catastrophic or Existential' Risks — Benzinga · U.S. retail-trader financial news site; commercial, center
- Anthropic IPO Prospectus Warns Its AI Could Pose 'Existential Risks To Humanity' — Forbes · U.S. business magazine, majority-owned by Integrated Whale Media; center, pro-business
- Leaked Anthropic IPO Prospectus Shows $42BN Net Loss, $518BN In Unfunded Spending Commitments, And $20BN In Cash — ZeroHedge · Pseudonymous U.S. financial blog; right-libertarian, bearish on markets
- Anthropic Lost $42 Billion on $4.6 Billion in Revenues in 2025: Leaked IPO Prospectus — Wolf Street · Independent U.S. financial blog (Wolf Richter); consistently skeptical of bubbles
- Anthropic IPO Prospectus Leaked: $42B Loss, Over Half A Trillion In Cloud Commitments — Skeptics Lash Out — Stocktwits · Retail-trader social platform news desk; engagement-driven
- Anthropic's prospectus details losses, growth, and, yes, a warning that its AI could end humanity — TechCrunch · U.S. tech trade press owned by Yahoo (Apollo Global); tech-industry focus
- Anthropic's leaked IPO prospectus details steep losses, rapid growth, and a fear that AI could end humanity — Fortune · U.S. business magazine, privately owned; center, pro-business
- Anthropic warns of AI's 'existential risks to humanity' in IPO prospectus — Daily Sabah · Turkish daily aligned with the Erdoğan government