
Anthropic’s path to the stock market is bringing a harder question into view. Claude is growing quickly, but how much will it cost to keep building the models and computing capacity behind it? A confidential IPO prospectus reviewed by Reuters gives investors a first look at both sides of that equation: rapidly rising revenue and enormous long-term infrastructure obligations.
Reuters reports that Anthropic’s 2025 revenue grew roughly twelvefold to nearly $4.6 billion. The company nevertheless recorded an operating loss of more than $8 billion, excluding writedowns tied largely to earlier fundraising. Compute and infrastructure cost $7.33 billion last year, about three times the 2024 figure and more than half of its reported operating expenses.
The headline net loss of nearly $42 billion needs care. According to Reuters, roughly $34 billion of that total was an accounting charge reflecting the increased estimated value of financing that could convert into Anthropic shares. It was not $34 billion spent keeping Claude running. The operating loss and compute bill remain substantial, but treating the full net loss as a cash burn would misstate the economics investors are being asked to assess.
The prospectus also describes about $518 billion in cloud, compute and infrastructure obligations across coming years, Reuters says. That is a long-term commitment figure, not a bill due in one year. Even so, it illustrates the scale of capacity Anthropic believes it will need if demand for AI keeps growing. More customers could help fill that capacity; slower adoption or falling model prices would make those commitments harder to justify.
There is another concentration risk. Reuters says two customers accounted for nearly a quarter of Anthropic’s revenue in 2025, while some major clients are not bound by long-term contracts. The company ended the year with $20.28 billion in cash, equivalents and short-term investments. That gives it resources, but it does not remove the pressure to turn growth into durable margins.
This is the financial detail missing from much of the talk around an AI IPO. Earlier, Anthropic’s public-market pitch looked like a referendum on AI’s promise and risks. The numbers add a more practical test: can a leading AI lab earn enough from its products to support a compute plan of this size?
Anthropic has not publicly released the confidential prospectus, and Reuters says the company declined to comment. An IPO, its timing and any valuation are therefore still uncertain. What is clear from the reported figures is that strong sales growth and staggering investment requirements are advancing together. Public investors would have to decide whether that balance is a path to a lasting business or a warning about how expensive the AI race has become.







