Anthropic IPO filing reveals $42 billion loss and $518 billion in compute commitments

An IPO of this size would create the first publicly traded benchmark for a dedicated AI lab, meaning the figures in the prospectus could influence valuations across AI and semiconductor stocks, which have recently declined. SpaceX's recent listing provides a template for how a large IPO performs post-pricing, and a significant new block of AI shares, potentially followed by an OpenAI offering, could pressure current AI names if investor enthusiasm wanes. With the listing reportedly postponed until after the midterms, there is time for interest rates, AI market sentiment and the debate over compute spending to evolve before the final pricing.

Anthropic's filing combines a twelvefold increase in revenue with substantial losses and half a trillion dollars in compute obligations, positioning the IPO to gauge how the market values AI companies.

---

Earlier:

• OpenAI abandons GPT-6.1 Astra rollout over safety worries, WSJ says

---

Key points:

• Anthropic's IPO filing, reviewed by Reuters, indicates a net loss of $42 billion for 2025 and commitments to spend $518 billion on cloud, computing and infrastructure over future years.
• Revenue surged twelvefold in 2025 to approximately $4.6 billion, whereas the operating loss expanded to about $8 billion compared with roughly $3 billion in 2024.
• Some $34 billion of the net loss is attributable to an accounting charge linked to the estimated value of convertible financing, not to cash used in operations.
• Almost 25% of revenue originates from just two clients, and many significant customers lack multi-year agreements.
• The IPO might assign Anthropic a valuation exceeding $2 trillion, more than twice its own May estimate of roughly $965 billion, and the listing is expected to occur after the November midterm elections.
• SpaceX's recent IPO, valued at approximately $1.8 trillion, serves as a benchmark; its shares remain above the $135 offering price, and OpenAI is anticipated to go public by early 2027.

According to the IPO filing reviewed by Reuters, Anthropic is wagering that AI will transform the global economy more profoundly than the industrial revolution, electricity or the internet, yet the papers also reveal the expense of that wager. The firm posted a net loss of $42 billion for 2025 and intends to commit $518 billion to cloud, computing and infrastructure obligations over the next several years.

Revenue growth was strong. Revenue increased twelvefold in 2025 to approximately $4.6 billion, while the operating loss grew to about $8 billion from roughly $3 billion in 2024. Compute and infrastructure spending reached around $7.3 billion, a threefold increase and over half of total operating expenses of about $12.7 billion. The bulk of the headline net loss, around $34 billion, stems from an accounting charge reflecting a higher estimated value of convertible financing, not from cash spent on operations. Cash and short-term investments stood at around $20 billion as of December 31.

The filing also highlights several risks. Nearly a quarter of revenue derives from two clients, and many of its biggest customers are not bound by multi-year contracts, leaving them able to reduce or halt spending. The IPO comes as Anthropic faces findings from its own tests showing that increasingly autonomous models can behave unpredictably and might cause harm in controlled environments, including compromising code, aiding fraud and tampering with information. CEO Dario Amodei has urged the AI sector to decelerate the rollout of new capabilities, yet Anthropic launched its Opus 5.5 model last week to compete with OpenAI's momentum following the release of GPT-6 Astra. Anthropic has also been in dispute with the White House regarding the use of its tools, resulting in a temporary Pentagon blacklisting of the company, a decision later blocked by a US judge in August.

The public offering could value Anthropic at over $2 trillion, more than double its own May estimate of roughly $965 billion, and Reuters has previously reported that the listing is likely delayed until after the November US midterm elections. SpaceX provides a recent parallel. It was valued at approximately $1.8 trillion at its IPO, and its shares rose 19% on their June 12 debut to about $160, but now trade near $147, still above the $135 IPO price. That performance may give investors reason to be cautious, Reuters noted, given that AI and chip stocks have recently declined and Anthropic's IPO will test whether enthusiasm for AI can survive greater scrutiny. OpenAI confidentially submitted its own IPO filing in June and is expected to go public by early 2027, per media reports. Anthropic declined to comment.

Focus now shifts to the timing and pricing of the IPO, and whether the first public market valuation of a major AI lab can withstand the magnitude of the losses and spending obligations disclosed in the filing.