Anthropic filed its S-1 publicly on October 26 and plans to list on Nasdaq as ANT. Fortune reports(opens in a new tab) first half 2026 revenue of $16.3 billion, up from $1.2 billion a year earlier, a GAAP net loss of $80.7 billion, and positive operating cash flow of $3.9 billion.
Read past the loss before you read it as a vendor risk. Fortune says about $72.4 billion of it is an accounting charge on convertible notes. The operating loss was $693 million, and adjusted operating income was $1.6 billion. On the numbers, this is a vendor that collects cash, not one about to disappear. The harder risks sit elsewhere in the filing.
The first is how you pay. Yahoo Finance's read of the filing(opens in a new tab) puts consumption revenue at 80% of the half and subscriptions at 20%. Your model bill is metered, and the vendor's growth story depends on that meter running faster. Agents that loop, retry and call tools are exactly how it runs faster. Set per team spending caps and alerts now, and ask finance to forecast AI cost as a usage line with a range, not a fixed license.
The second is bargaining power. Fortune says the top five customers made up about a quarter of revenue in the first nine months, and nearly half of revenue flows through the Amazon, Google and Microsoft marketplaces. If you already hold a cloud spending commitment, buying model usage through it can draw down that commitment instead of opening a new contract. A large buyer that can move volume between vendors has more pricing room than the list price suggests.
The third is commitment. Yahoo counts about $430 billion in noncancelable compute commitments to Broadcom, Google, Amazon and Microsoft. A lab carrying that much fixed cost needs volume, which helps buyers on price but also means terms can change when capacity is tight. Fortune also notes that 5% to 7% of this year's revenue came from customers later found to violate usage policies. Have counsel confirm your use cases sit inside the acceptable use policy, because the filing shows the vendor finds violators even among paying customers.
Action items
Anthropic's public filing shows 80% of its revenue is metered usage. That makes your AI bill a variable cost the vendor wants to grow.
For the CAIO
Put spending caps on every team using models by the token, and buy through your existing cloud commitments where you can.
Researched and drafted by an automated workflow, then reviewed and edited by a human editor before publication. Every source is linked. See how we use AI here.
Also worth knowing
- FortuneAnthropic's S-1 IPO prospectus shows $16.3 billion in revenue and an $80.7 billion net loss(opens in a new tab)
Most of the $80.7 billion loss is a $72.4 billion convertible note charge; operating cash flow was a positive $3.9 billion. Judge vendor viability on cash, and budget for metered usage that rises with agent adoption.
- Reuters via Honolulu Star-AdvertiserHow one hacker used AI agents to break into over 100 company websites(opens in a new tab)
Using the open source Cairn platform, one operator's agents tried 1,200 sites and breached over 100, including sites tied to Deere, SharkNinja and Cummins, for about $50 in tokens each. Audit vendor run and support sites first.
- Google CloudGemini Enterprise release notes(opens in a new tab)
From October 26, AlphaEvolve experiments with no model named default to Gemini 4 Argon at its token prices, and admins can now cap per user quota. Pin models in code so a vendor default change cannot raise your bill.
- FinTech GlobalAI deployment gap exposes uneven progress across financial institutions(opens in a new tab)
A RegTech Analyst index scores risk and compliance AI at 51.9 of 100, with decision impact at 62.4 but autonomy at 44.2, and a 36.5 point gap between leaders and laggards. Benchmark your own compliance stack against it.