What do we know about Anthropic’s finances?
Anthropic is still private, so we cannot see normal public financial statements. Most of what we know comes from fundraising materials, people who have seen them and the numbers Anthropic chooses to announce. This makes the picture much less complete than it looks.
What has actually happened?
| Q1 2026 revenue | $4.8B |
|---|---|
| Annualized revenue in early May | over $47B |
| Latest funding round | $65B |
| Valuation after the round | $965B |
The $4.8B is reported revenue for a finished quarter. The $47B is a run rate. It means that if Anthropic kept earning money at the same pace for a year, revenue would reach around $47B. It does not mean Anthropic had already earned $47B.
Sources: Reuters, based on Anthropic’s fundraising materials, and Anthropic’s Series H announcement.
What was projected?
| Q2 2026 revenue | $10.9B |
|---|---|
| Q2 2026 operating profit | $559M |
These were projections in fundraising materials, not the result of a finished quarter. If they became real, Anthropic would have reached its first operating profit. The projected margin was only around 5%, but this would still matter a lot for our question. It would show that a frontier lab can cover very large costs through normal model and product revenue, at least for some time.
It would not yet show that the model works after competitors catch up, or that it can pay for every new round of frontier training.
How much is it committing to compute?
Anthropic reportedly agreed to pay SpaceX $1.25B every month through May 2029 for compute capacity. That is around $15B a year if the monthly amount stays the same.
This is not the same as training cost. The compute can be used both to train models and to serve customers. It is also a commitment, not proof that the same amount has already been spent.
Source: Reuters.
Who is funding it?
Anthropic raised $65B in May 2026 at a $965B valuation. The round included $15B that had already been committed by large cloud companies, including $5B from Amazon.
Other investors included Blackstone, Brookfield, Coatue, Dragoneer, Fidelity, GIC, Goldman Sachs, Lightspeed, Sequoia, T. Rowe Price and Temasek. Anthropic had already raised large rounds before this, while Amazon separately said its total investment had reached $8B in 2024.
Sources: Anthropic’s Series H announcement and its Amazon partnership announcement.
What this tells us
Anthropic might be the strongest argument against the simple version of my prediction. Its revenue is growing insanely fast, and it believed it was close to an operating profit even while paying for a huge amount of compute.
But I don’t think this answers the main question yet. We still need to know how much of this revenue comes from Anthropic having the best model, how long that advantage lasts and what happens to demand when a much cheaper model becomes good enough.
What we still do not know
- Whether Anthropic actually reached the projected Q2 profit.
- Its gross margin and the direct cost of serving Claude.
- How much it spends on training frontier models.
- How much of the compute commitment is for training and how much is for serving customers.
- How much revenue comes from Claude Code, the API, subscriptions and large enterprise contracts.
- How concentrated the revenue is among a small number of customers.
- Whether it can remain profitable while continuing to push the frontier.