The reported filing puts a scary loss up front. For Canadian firms building on Claude, the commitments and customer mix matter more.
Anthropic’s IPO filing, as reported by Reuters on 28 September 2026, shows a 2025 net loss of about $42 billion on revenue of roughly $4.6 billion (all figures US dollars). That headline loss is mostly an accounting charge. About $34 billion came from revaluing financing instruments, which leaves an operating loss near $8.06 billion. The number an Anthropic IPO reader who buys Claude should sit with is a different one: roughly $414 billion in compute commitments the company cannot cancel.
One caution before the numbers. We searched SEC EDGAR on 30 September and found no public S-1, so every figure below is an attributed claim from reporting on a draft, not a document we have read.
TL;DR
- Reported 2025 results: $4.6 billion in revenue and a $42 billion net loss, of which about $34 billion is non-cash.
- Anthropic has $518 billion in cloud and infrastructure commitments across six partners, about 80% of them non-cancellable.
- Two unnamed customers made up nearly a quarter of 2025 revenue.
- Our math: the locked commitments equal about 6.4 years of the reported July 2026 run rate.
- For buyers, the risk sits in pricing and priorities, not in Claude disappearing.
What did Anthropic’s IPO filing reveal?
According to reporting on the draft, Anthropic grew revenue from about $400 million in 2024 to $4.6 billion in 2025, then booked $11.5 billion in the second quarter of 2026 alone, with an annualised run rate of $65 billion by July. It also listed $518 billion in cloud and infrastructure commitments, plus a founder structure that keeps 50.1% of voting power.
The document reportedly runs 261 pages, and about 80 of them are risk factors. Reuters reports the target valuation is above $2 trillion. That is a lot of confidence riding on one quarter’s growth.
Why it matters
Most Canadian firms that use AI now buy it from a handful of labs, and Anthropic is one of the biggest. A supplier’s finances shape its pricing, its product priorities and how hard it leans on its largest accounts. This filing is the first look at those pressures from the inside, even second-hand, and it lands while many owners are picking a default model.
Here is our own arithmetic, and it is analysis rather than reported fact. $414 billion of non-cancellable commitments against a $65 billion run rate is about 6.4 years of revenue already promised to cloud and chip partners. Against 2025 revenue, the full $518 billion is about 113 times sales. Anthropic can grow into that. It has to.
Then there is the customer mix. Two customers produced nearly a quarter of 2025 revenue, with no long-term contracts binding them. If you spend a few hundred dollars a month, you are not who the roadmap is built around. Nobody is.
What does this mean for businesses that use Claude?
For a firm spending modestly on Claude, nothing changes this week. The pressure shows up later, in three places: price changes on the models you depend on, model retirements on the lab’s schedule rather than yours, and features built for the largest accounts first. Treat Anthropic like any supplier with heavy fixed costs and plan around it.
None of this is unique to Anthropic. Every frontier lab is signing compute deals of this size. Anthropic is simply the first to show its bill in a securities document.
What should leaders do next?
- List every workflow that calls Claude, directly or through a vendor. Plenty of SaaS tools embed it without saying so on the invoice.
- Test one alternative model on your most important workflow this quarter. Our take on why ChatGPT vs Claude vs Gemini is the wrong first question sets out how to run that test.
- Ask each AI vendor what notice you get on price changes and model retirements, and get the answer in the contract. The roadmap contract problem covers the clauses.
- Keep your prompts, test cases and source data in your own storage, not only inside a vendor’s project space.
- Hold off on annual prepay for any single model until the public S-1 is out and you can read the risk section yourself.
The sceptic’s view
The strongest counter-argument is simple. The loss is mostly non-cash, revenue grew more than tenfold in a year, and one quarter of 2026 brought in two and a half times all of 2025. Compute commitments are how every lab secures chips, and a customer list with giants on it is validation, not weakness. On that reading, Anthropic is the safest bet in the category.
That may prove right. The buyer steps above still cost almost nothing, and they pay off whichever way it goes.
What to watch
- The public S-1 on SEC EDGAR, which US rules require at least 15 days before an IPO roadshow. Check whether the two large customers are named.
- Any Claude API price or plan changes before 31 December 2026.
- Whether the six compute partners and their commitment sizes are disclosed in the public version.
FAQ
Is Anthropic going public?
Reuters reports that Anthropic has prepared an IPO prospectus targeting a valuation above $2 trillion. As of 30 September 2026 no public S-1 appeared on SEC EDGAR, so timing is unconfirmed.
Did Anthropic really lose $42 billion in 2025?
That is the reported net loss. About $34 billion of it came from revaluing financing instruments, a non-cash charge, which leaves a reported operating loss of about $8.06 billion.
Should Canadian businesses stop using Claude?
No. The filing points to pricing and priority pressure, not to Claude going away. Keep using what works, and keep a tested second model for your most important workflow.
Closing analysis
The loss grabbed the headline, but the commitments write the future. A company that owes its suppliers six years of revenue will price, ship and prioritise to meet that bill, and every customer outside the top two should plan for it. More in our AI in business coverage.
Written by the AI Magazine Canada team, reviewed by the AI Magazine Canada editorial team.
Disclosure
This piece was drafted with AI tools, including Anthropic’s Claude. The author has no relevant financial, advisory, or board relationships with any party named in this column.
