Realtime AI News
Anthropic's IPO Filing Shows $17B Revenue—and Compute Eating 58% of It
Anthropic's updated IPO filing disclosed quarterly revenue of $17 billion, up from $11.5 billion in the prior quarter, while compute costs climbed to 58% of revenue from 48%. The operating loss widened, putting unit economics at the center of the company's listing story.

Anthropic is finally giving investors a clearer view of the economics behind its rapid growth. In an update to its IPO filing reported by TechCrunch on November 3, the company extended its financial disclosures from June 30 to September 30, adding a full quarter of operating data.
Top-line growth remains steep. Revenue reached $17 billion for the quarter, up from $11.5 billion in the previous one—a pace close to 50% quarter over quarter, which the filing presents as evidence of durable enterprise demand.
The bottom line tells a different story. The operating loss grew alongside revenue, driven largely by compute spending. Compute consumed 58% of revenue in the quarter, up from 48% the quarter before—moving the wrong way at the moment Anthropic most needs to show IPO investors that its unit economics are improving.
The filing points to procurement structure as the culprit. Anthropic's compute bill hit roughly $10 billion for the quarter, including $5.0 billion of capacity bought under short-term commitments, which the document says “currently costs significantly more than comparable compute procured under our multi-year commitments.” Demand simply outran contracted supply.
A chart in the filing makes the squeeze visible: revenue per dollar of compute climbed steadily from the start of 2025, peaked at just over 2x this quarter, then slipped back to 1.72x as expensive stopgap capacity diluted the returns.
Those stopgap deals are large. TechCrunch notes agreements with so-called neoclouds—smaller, more flexible providers that charge higher rates than hyperscalers—including a commitment of $1.25 billion a month for compute alongside multi-billion-dollar deals with several vendors.
Anthropic says it is working to bring costs down through an in-house effort it calls “Anthropic Direct.” But with demand growing at this pace, owned capacity cannot keep up, and the filing states that compute and infrastructure expenses will “remain our largest operating expense for the foreseeable future.”
The bigger risk is structural. As long as revenue keeps compounding at nearly 50% a quarter, few investors will punish high compute costs; if the data center buildout slows because bond markets turn skittish or local opposition hardens, those costs become much harder to absorb—and there is little the company can do about either.
For Anthropic, the IPO will price two things at once: whether revenue keeps compounding, and whether compute turns from a drag into an advantage.
Why it matters
The filing sets a benchmark for how much AI revenue the infrastructure buildout absorbs, and how exposed lab margins are to compute prices ahead of Anthropic's listing.
Nearby Updates
All11/04, 01:03
Alibaba Cloud Brings AI and Cloud Infrastructure to Dakar 2026 Youth Olympics
Alibaba is supplying cloud computing and AI technology for the Dakar 2026 Youth Olympic Games, the first edition of the event held in Africa. The deployment puts Alibaba Cloud's sports-technology stack into a high-pressure, low-tolerance environment far from its home market.
11/04, 00:56
Tokens Land on Double 11 Shelves: Prices as Low as 1 Cent, 2 Trillion Sold in 30 Minutes
AI services became a headline category in this year's Double 11 shopping festival, with token packages priced as low as one cent and one livestream selling 2 trillion tokens in half an hour, according to National Business Daily. The real test is retention: cheap tokens pull in first-time buyers, but converting them into paying users is a different problem.
11/04, 00:43
Chargebacks911 sells merchants cover for chargebacks on AI agent purchases
Chargebacks911 has launched an AI agent purchase indemnification product that identifies agent-initiated transactions and covers qualifying chargebacks for merchants. The capability is bolt-on, so merchants keep their existing fraud prevention stack and payments infrastructure, and the company says it applies across verticals and markets.
11/04, 00:35
IBM's Palette splits decks into a designer and a coder, and a 20B model keeps up
IBM Research has introduced Palette, a 20B open-weight model fine-tuned to build slide decks in a specific design language by separating a designer that writes a JSON brief from a coder that generates each slide. On PresentBench it ranks first in the Economics and Education domains and second overall at 59.7 points, behind NotebookLM's 60.8.