Markets · September 29, 2026 · 12 min read
Reuters saw Anthropic’s IPO prospectus on September 28 and printed nearly $4.6 billion in 2025 revenue against more than $8 billion in operating loss, while a $42 billion net loss mostly reflects financing marked up on paper, not cash spent on servers.

On September 28, Reuters published an exclusive that begins with a document the public still cannot open on EDGAR. Echo Wang in New York and Manya Saini in Bengaluru reported that they had seen Anthropic’s IPO prospectus. The numbers in the story, Reuters said, were appearing “here for the first time.” Anthropic declined to comment. The company had confidentially submitted a draft S-1 on June 1, 2026, and said then that the share count and price were not set. As of this week, the registration statement Reuters describes has not replaced that silence with a public filing anyone can download.
That gap matters for how to read the piece. Every figure below comes from Reuters’s account of a prospectus it saw, not from a Form S-1 on the SEC website. Secondary outlets, including CNBC and The Hindu BusinessLine, republished the same Reuters report. When the story says Anthropic lost $42 billion in 2025, it also says more than $8 billion of that story is operating loss, with writedowns tied to earlier fundraising stripped out. When it says the company plans to spend $518 billion on cloud, computing, and infrastructure obligations in coming years, it is quoting risk and commitment language from a document written for investors, not a check Anthropic has already cut.
A bet bigger than electricity
Reuters opens with the prospectus’s own framing. Anthropic, in the document, is making a massive bet that artificial intelligence will transform the global economy more profoundly than industrialization, electricity, and the internet. That sentence is ambition on letterhead. The next numbers are the bill. Anthropic reported a net loss of $42 billion in 2025, according to Reuters, and plans to spend $518 billion on cloud, computing, and infrastructure obligations in coming years.
Revenue, in the same document, grew sharply. Reuters said 2025 revenue rose twelvefold to nearly $4.6 billion. TradingKey and other outlets that cite the same Reuters piece sometimes round that figure to $4.59 billion. The operating loss widened with the growth. Reuters put the 2025 operating loss at more than $8 billion, excluding writedowns of liabilities mostly tied to previous fundraising. The same reporting chain, in outlets that preserved the full Reuters table, puts the operating loss at $8.06 billion in 2025 against $2.98 billion in 2024. Revenue in 2024, in that reconstruction, was about $386 million. The lab is five years old in public-market terms and already spending like an infrastructure company.
Compute is where the money goes. The AI lab spent $7.33 billion on compute and infrastructure in 2025, Reuters reported, a threefold increase from 2024. That line was more than half of $12.65 billion in total operating expenses. Spend on tokens and racks ran about 1.6 times the year’s revenue in that accounting. That ratio is the structural fact behind the IPO story. A software company that loses money while scaling often does so on people and sales. Anthropic’s prospectus, as Reuters read it, centers the cost base on purchased capacity from cloud partners who are also investors.
The $34 billion that is not a server bill
The headline net loss scares faster than the operating loss because $42 billion sounds like cash on fire. Reuters is explicit that it is not, mostly. The near-$42 billion net loss included a roughly $34 billion accounting charge. That charge reflected an increase in the estimated value of financing that could eventually turn into Anthropic shares, rather than money the company spent running its business. GAAP net loss and operating loss diverge when prior rounds carry instruments that get remeasured as the company’s value rises.
Reuters still treats the operating loss as real. More than $8 billion of operating red ink on nearly $4.6 billion of revenue is not a paper entry. It is the cost of training, serving, and staffing a frontier lab while prices compress and rivals ship weekly. The prospectus, in Reuters’s telling, also shows cash on the balance sheet that is large until you set it beside the forward commitments. Anthropic had cash, cash equivalents, and short-term investments totaling $20.28 billion as of December 31, the documents show. Twenty billion dollars is a fortress until the same filing describes $518 billion in cloud, computing, and infrastructure obligations stretching into coming years. Reuters does not, in the September 28 story, print the schedule of those payments. The asymmetry is the point. One side of the business is signed for years. The other side is customers who can leave.
Two customers, a quarter of revenue
Reuters reported that nearly a quarter of Anthropic’s revenue last year came from two customers. The prospectus, in the risk-factor section Reuters summarized, warned that many of the largest clients were not locked into long-term contracts and could cut or stop spending. That paragraph is standard IPO language sharpened by concentration. When two unnamed accounts can move the top line by a quarter, the twelvefold revenue growth story carries a counterweight. Public investors, if the offering happens, would be buying into a company whose biggest buyers can walk away on short notice while Anthropic still owes cloud vendors on long horizons.
Amazon and Google appear in the Reuters story as early strategic partners who invested billions while supplying the cloud infrastructure used to train and deploy Claude models. The September 28 piece does not print ownership percentages. Other reporting, not in the Reuters exclusive itself, has described Amazon and Alphabet as large shareholders. Reuters keeps the partner names at the level of the prospectus narrative: both companies fund the startup and sell it capacity. That dual role is familiar in AI. It also means the IPO roadshow, when it comes, will ask whether revenue is arm’s-length or circular.
Reuters illustrated the corporate story with a photograph from eleven days earlier. People visit an Anthropic booth at Moscone Center during the Dreamforce 2026 technology summit in San Francisco, California, September 17, 2026. The credit is Carlos Barria/Reuters. The IPO article uses that file photo under the valuation section. Afterimage is not republishing Barria’s image here. Reuters marks its pictures for licensing, and the booth scene is available on Reuters’s site with the September 28 exclusive. The hero on this page is separate signage supplied for Afterimage. The Dreamforce booth is the public-facing version of the same company the prospectus describes as racing toward a listing.
More than $2 trillion on the cover
Reuters said the public sale could value Anthropic at more than $2 trillion. That target is more than double an estimated valuation of $965 billion in May, in the same reporting. No share count, no price range, and no proceeds goal appear in the September 28 story. Those blanks are consistent with Anthropic’s June 1 announcement that the number of shares and the price had not yet been determined.
The listing would cap a year of large US IPOs, Reuters noted, despite elevated interest rates and valuation worries. It would also follow SpaceX’s debut, which Reuters used as a comparator. SpaceX’s recent IPO valued Elon Musk’s company at $1.77 trillion. Shares surged 19 percent on the June 12 debut to $160, Reuters wrote, and traded around $147 afterward, above the $135 offer price. That path is both encouragement and warning. SpaceX entered public markets with a business investors could touch. Anthropic enters, if it enters, as a pure-play bet on frontier models while AI and chip stocks have sold off recently. Reuters said Anthropic’s sale will test whether enthusiasm for the AI trade can withstand greater scrutiny, given ambitious growth projections in the prospectus.
The May valuation Reuters printed, $965 billion, and the IPO talk above $2 trillion, are not the same kind of number. The first is a private-markets estimate attached to a fundraising moment. The second is what Reuters said the public sale could seek. Doubling in four months, if the listing price matches the talk, would tell you how much public demand the underwriters expect for a pure-play frontier lab. It would also tell you how much skepticism a miss would invite if AI equities keep selling off. Reuters explicitly tied recent weakness in AI and chip stocks to the test Anthropic’s sale would pose.