Anthropic’s IPO paperwork shows a $42 billion loss and a $518 billion bill still to pay

Anthropic lost $42 billion last year and has signed up to spend $518 billion on cloud, computing and infrastructure, according to the company’s IPO prospectus. Reuters has seen the document and reported its first details on Monday. It isn’t public yet: Anthropic hasn’t filed it openly with the US Securities and Exchange Commission.

The numbers

The prospectus shows a company growing extremely fast while spending even faster, according to Reuters:

• Revenue: up 12-fold in 2025, to nearly $4.6 billion.
• Losses: a net loss of about $42 billion, and an operating loss of more than $8 billion.
• Compute: $7.33 billion spent on compute and infrastructure in 2025, three times 2024’s figure and more than half of its $12.65 billion in operating costs.
• Commitments: $518 billion in cloud, computing and infrastructure obligations to come.
• Cash: $20.28 billion in cash, cash equivalents and short-term investments at the end of 2025.

The headline loss is less dramatic than it looks. About $34 billion of it is an accounting charge that reflects the rising value of financing that could later turn into Anthropic shares, rather than money spent running the business.

What investors will worry about

The risk factors show how concentrated Anthropic’s business still is. Nearly a quarter of its 2025 revenue came from just two customers, and it warns that many of its biggest clients aren’t locked into long-term contracts and could cut back or stop spending.

The prospectus also pitches a sweeping vision: that AI will change the global economy more deeply than industrialisation, electricity or the internet. The listing could value Anthropic at more than $2 trillion, more than double its own estimated valuation of $965 billion in May, and would probably come after the US midterm elections in November, Reuters reports.

Who runs it

The document lists the leadership investors would be buying into, including co-founders Dario and Daniela Amodei, finance chief Krishna Rao and chief technology officer Rahul Patil. It also names the trustees of Anthropic’s Long-Term Benefit Trust, the body set up to keep the company focused on safety, which now includes former Federal Reserve chair Ben Bernanke, according to a Reuters factbox.

The timing is awkward. Amodei has called on the industry to slow the pace of new releases, and Anthropic’s own research keeps finding that autonomous models can behave in harmful ways in tests. Meanwhile, the company is still fighting the Pentagon’s supply-chain blacklist, and Amodei dined with President Trump on Sunday night. It has also been locking in computing power, most recently in an $11.6 billion deal with Akamai.

Anthropic hasn’t commented on the prospectus. OpenAI confidentially filed for its own IPO in June, Reuters says, so the race to list is on.

Update: an extinction warning, and the founders keep control

Reuters has since reported two more details from the prospectus. The filing warns investors that advanced AI could pose “catastrophic or existential risks to humanity,” and that its models could show “self-preserving behaviors,” including attempts to “resist shutdown,” to “conceal or manipulate information” and behaviour “resembling blackmail.” About 80 pages of the 261-page main body are risk factors, nearly twice the 48 pages describing the business. It also admits a problem safety researchers have raised for months: “Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.” It comes days after Anthropic safety researcher Evan Hubinger put the chance of AI killing all humans within a decade at more than 10%.

Control will stay with the people who started it. A new “Founder LLC” made up of Anthropic’s seven co-founders will hold a single Class F share carrying 50.1% of the total voting power on key matters, including electing some board members, according to the filing. Anthropic will stay a public benefit corporation, and the Long-Term Benefit Trust will elect four of the board’s directors. The company warns that this setup could lead to decisions “that may conflict with short-, medium-, or long-term financial interests.” It also says it has chosen not to build image and video generation models so it can spend its computing power on research and safety.

Update: SpaceX, Amazon and Google, and the risk of rogue agents

Reuters has reported three more details from the confidential filing. The first is the size of Anthropic’s computing deal with Elon Musk’s SpaceX: agreements worth up to $84.5 billion through 2029, almost double the roughly $45 billion SpaceX’s own filings had suggested. Unlike most of the $518 billion Anthropic has committed to, around 80% of which it must pay whatever it uses, the SpaceX agreements can generally be ended on 90 days’ notice.

The second is how much Anthropic depends on its biggest backers. It routed 47% of last year’s sales through Amazon’s and Google’s cloud marketplaces, up from 11% in 2023, and paid them roughly $351 million in fees, according to Reuters’ analysis. Both companies are also investors, suppliers of computing power and direct rivals. The filing admits this “creates complex dynamics that could give rise to conflicts of interest and adversely affect our access to compute.” Two unnamed customers each brought in 12% of revenue.

The third is the legal risk from its own AI agents. The filing warns that errors, misalignment or security exploits in autonomous agents “may result in real-world consequences”, that how existing laws apply to AI agents is “unsettled”, and that the liability limits in its contracts may not be “enforceable or adequate”. The same question is now being tested in court: on Tuesday, a nonprofit sued OpenAI over its agents’ Hugging Face hack, arguing that “the AI did it” is no defence.

Update: Broadcom will lend Anthropic up to $42 billion

The chipmaker Broadcom has agreed to lend Anthropic up to $42 billion to lease its chips and pay for infrastructure, according to the prospectus, Reuters reported on Thursday. The money comes as a convertible note and could cover about a third of the $125.2 billion Anthropic has committed over five years to lease capacity on Google’s tensor processing units, the AI chips Broadcom helps design. Anthropic is expected to become the biggest customer of Broadcom’s chip design business next year.

That makes Broadcom a supplier, a landlord and a lender at once. The filing says Anthropic put cash into a restricted account for Broadcom’s benefit in April and may have to deposit more, and it warns of “potential conflicts of interest” that could affect its access to computing. If Anthropic missed a payment, it could have to pay its lease obligations immediately and lose access to the loan. It is the same pattern as Amazon and Google: the companies Anthropic relies on for computing are also financing it.

Broadcom’s banks have now started gathering about $60 billion of financing to fund chips for Anthropic and other AI companies, Bloomberg reported on Friday, and are preparing to send out syndication letters for a $42 billion senior-secured tranche. Broadcom and Anthropic haven’t commented on the report.

Update: Anthropic warns Washington could cost it customers

The filing also warns that the US government’s attitude to Anthropic could hurt it well beyond its government work, Reuters reported on Friday after seeing the prospectus. Government contracts bring in less than 1% of Anthropic’s annual revenue, but the company says government actions could cause “material revenue losses or business disruptions” and lead to “adverse media coverage, public scrutiny and negative perceptions among existing and potential customers, partners, employees and investors”.

It gives examples from the past year: the February order for federal agencies to stop using its models, when the Pentagon also named Anthropic a supply-chain risk, and the Commerce Department’s worldwide export restrictions on Fable 5 and Mythos 5 in June, which forced it to switch the models off temporarily for all customers to comply. The warning comes days after Trump hosted Dario Amodei at the White House. Anthropic hasn’t commented on the report.

Update: Dario Amodei made $18 million last year, and the founders pledge 80% of their shares to charity

Dario Amodei was paid $18 million as Anthropic’s chief executive in 2025, according to the IPO filing, Reuters reported on Tuesday. Most of it came as stock and options. His sister, president Daniela Amodei, received $16.4 million. Both had their salaries doubled to $1.4 million in July, and the board has since promised them restricted stock units, some tied to staying at the company and some to the IPO. Finance chief Krishna Rao earned $720,250.

That puts Amodei above Alphabet’s and Amazon’s chief executives on the figures their companies report each year, but behind those at Oracle and Nvidia. The filing doesn’t show how much of Anthropic the founders own. It does say the Amodeis and their fellow co-founders have pledged 80% of their personal Anthropic equity to charitable causes. Anthropic declined to comment on the pay figures.

Update: New York City’s Comptroller asks Anthropic to pause its IPO

New York City Comptroller Mark Levine has urged Anthropic to pause its planned IPO, in a letter to Dario Amodei dated October 14. Levine is investment adviser, custodian and a trustee of the city’s pension funds, which hold over $325 billion in assets.

He argues that the company should first address the safety, governance and development-pace concerns it has raised itself, pointing to Amodei’s September 12 essay “We Must Pace the Frontier” and the company’s own risk warnings. He also says the “morally binding” White House accord Anthropic signed is “insufficient to meet the moment”.

The demands of preparing for a public offering diverts resources and attention from this very necessary work.

Mark Levine, New York City Comptroller

The letter also raises the governance structure Anthropic reportedly plans to adopt, which could give founders and trustees concentrated control, calling the gap between control and accountability “especially concerning for investors”. Levine asks for a written response and says he welcomes a discussion. Anthropic hasn’t publicly responded.

Update: a Harvard physicist joins the trust that picks Anthropic’s board

Christopher Stubbs, an experimental physicist and former Dean of Science at Harvard, has joined Anthropic’s Long-Term Benefit Trust, the body named in the prospectus that has the power to appoint a majority of the company’s board, Anthropic announced on October 21.

Stubbs was part of the High-Z Supernova Search Team that found the universe’s expansion is speeding up, and he has advised the US government on national security and arms control as a member of the JASON group. He joins chair Neil Buddy Shah, national security expert Richard Fontaine and former Federal Reserve chair Ben Bernanke. Anthropic says trustees are independent of its management and investors.

Stubbs said he uses AI tools in his research every day, and gave a personal reason for joining:

As a grandfather, I want to help ensure that AI is developed and deployed for the long-term benefit of humanity.

Christopher Stubbs, Long-Term Benefit Trust

Why it matters

This is the first detailed look at the finances of a frontier AI lab, and it shows the scale of the bet: billions in revenue, but hundreds of billions promised for computing. If investors buy in at a $2 trillion valuation, it will set the price of the whole AI boom.

On October 14, Anthropic told IPO investors its third-quarter revenue rose about 45% to $16.8 billion, and that it was profitable on an adjusted basis, according to reports of the meeting.

An updated draft of the prospectus shows $16.3 billion in revenue and an $80.7 billion net loss for the first half of 2026, most of it non-cash accounting charges, The Wall Street Journal reported on October 20.

On October 26 Anthropic filed its S-1 publicly, applying to list on the Nasdaq under the ticker ANT. The audited figures confirm the 2025 numbers above: revenue of $4.59 billion and a net loss of $41.97 billion.

Sources: Reuters (primary), Reuters factbox, Reuters on the risk factors, Reuters on the Founder LLC, Reuters on Big Tech partners, Reuters on rogue agents, Reuters on Broadcom, Bloomberg on Broadcom’s financing, Reuters on government attitudes, Reuters on executive pay, Anthropic on Christopher Stubbs.