Anthropic wants stock market investors to value it at more than $2 trillion. Last year, the company behind Claude AI lost $42 billion.
Both numbers sit in Anthropic’s IPO prospectus, the document a company files before selling shares to the public. Reuters reviewed it on Monday. Here are eight places where the pitch and the numbers pull apart.
1. A $2 Trillion Price Tag on $4.6 Billion in Sales
Anthropic made about $4.6 billion in 2025 revenue. At $2 trillion, investors would be paying roughly $435 for every $1 of sales.
Annualized revenue reached $65 billion in July and could hit $100 billion by year-end. But that is based on the current run rate, not a full year of audited sales.
That makes the valuation unusually dependent on growth continuing almost perfectly. A slowdown would make the $2 trillion figure look much harder to justify.
2. The Business Lost $8 Billion
Anthropic lost $8.06 billion from operations, up from $2.98 billion in 2024.
The headline $42 billion loss exaggerates the cash burn. Yet the underlying business is still losing billions, and those losses are growing.
The business is improving at speed, but it still loses money on every dollar of sales.
3. Anthropic’s Profit Claim Leaves Out the Cost of Building AI
Anthropic reported positive adjusted operating income in Q2.
That number excludes model training, partner revenue sharing, and stock-based pay.
This is where the profit claim gets awkward. Training future Claude models is central to Anthropic’s business, yet the adjusted figure removes that cost.
4. Anthropic Spends More on Computing Than It Earns
Anthropic spent $7.33 billion on compute in 2025 against $4.6 billion in sales.
That works out to about $1.60 of computing for every $1 of revenue.
For a software company, that is an unusual cost structure. Anthropic must make AI dramatically cheaper to run, or keep raising prices and revenue faster than its infrastructure bill.
Costs bite on the customer side too. Uber burned through its entire 2026 AI budget by April after giving Claude to its engineers. Microsoft restricted internal Claude use over token costs, BeInCrypto reported in May.
5. $518 Billion of Future Bills
This is the largest number in the filing. Anthropic plans to spend $518 billion on cloud, computing, and infrastructure obligations in the coming years.
It ended 2025 with $20.28 billion in cash. Many of those commitments reportedly cannot easily be canceled.
The mismatch is huge. Anthropic is locking in hundreds of billions of dollars of spending before anyone knows whether AI demand will stay this strong.
6. Two Customers Bring in Nearly a Quarter of Revenue
Nearly a quarter of Anthropic’s 2025 revenue came from just two customers. The Reuters report did not name them.
Many major clients also lack long-term contracts.
That makes the growth story more fragile than the headline numbers suggest. Losing even one major customer could materially change the revenue picture.
Government work carries its own risk. A clash with the White House over how its tools are used led the Pentagon to briefly blacklist Anthropic. A US judge blocked that move in August.
7. Its Investors Are Also Its Suppliers
Amazon and Google invested billions in Anthropic. They also sell it cloud infrastructure.
That creates an unusual loop: investment money can return to the same companies through cloud spending.
There is nothing inherently improper about that. But it makes it harder to see how much of the AI boom is independent customer demand and how much is tied to Big Tech funding the ecosystem around itself.
8. Anthropic Warns Its Own Product Could Be Dangerous
Around 80 pages of the filing discuss risks. Just 48 describe the business.
Anthropic says some models resisted shutdown during testing.
“Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.”
Michael Burry put it more bluntly:
“IPOs need hype & puffery; ‘we are so awesome it could become dangerous’ is hype & puffery.”
The unusual part is that Anthropic is asking investors to fund rapid expansion while warning that the technology itself may become difficult to control. Few IPOs contain a risk section this central to the product being sold.
What Comes Next for the Anthropic IPO
The listing is likely to come after the November US midterm elections. That delay would let Anthropic show third-quarter results before it pitches investors on its roadshow.
The benchmark is already on the board. SpaceX listed at a $1.77 trillion valuation on June 12. Its shares closed their first day at $160 and now trade near $147, above the $135 offer price.
OpenAI filed confidentially in June and is expected to list by early 2027. Analysts expect whichever lab lists first to set the price benchmark for the whole AI industry.
Anthropic’s pitch is ultimately a bet that extraordinary growth will arrive before extraordinary costs catch up.
