Anthropic, the US company that develops generative artificial intelligence (AI) models, reportedly generated revenue of $16.8 billion (CHF 14.02 billion) in the third quarter. That is the figure it announced to investors, according to the Financial Times and Bloomberg, representing an increase of about 45% in three months. These unaudited figures were presented on Wednesday at a day reserved for institutional investors, at the company’s headquarters in San Francisco, according to both media outlets. Anthropic declined to comment.

Strong growth, but slower

Growth remains marked, but it is slowing. In the spring, revenue had more than doubled, rising from $4.8 billion in the first quarter to $11.6 billion in the second. The increase of about 45% announced for the third quarter is therefore well below the pace of the previous quarter.

To understand what is at stake, it is worth recalling how this business model works. Each generation of AI models requires more computing power, first to be trained, then to meet demand that exceeds available capacity. Anthropic rents most of this computing power from Amazon, Google or SpaceX.

Profitable on an adjusted basis

The company also says it is profitable for the second consecutive quarter, on an adjusted basis, according to the FT. This is a non-standard measure that notably excludes stock-based compensation. Anthropic also claims that it now generates more cash than it consumes, and that it brings in $1.70 of revenue for every dollar spent on computing power, according to the same sources.

Conventional accounting paints a different picture. According to documents seen by Bloomberg and Reuters, the company lost nearly $42 billion in 2025, of which about $34 billion consisted of accounting entries with no money actually spent. Its operating loss, closer to actual spending, exceeded $8 billion, on revenue of $4.6 billion in 2025.

These indicators cannot be verified at this stage, as no accounts have been published: the IPO prospectus, filed confidentially in June, has still not been made public. They should therefore be read as statements by the company, relayed by the press, and not as certified results.

Revenue far from the sums committed

Even with this sharp rise, this revenue remains far below the amounts committed to the AI race. According to the FT, Anthropic plans more than $500 billion in infrastructure spending over ten years. This is the whole bet of the sector: financing colossal computing capacity today while counting on demand that continues to grow.

An IPO expected in November

These figures come a few weeks before an IPO now expected in November, one that could set a record. According to Bloomberg, Anthropic could launch the sale of its shares as early as the week of 9 November, for a listing before Thanksgiving, on 26 November. This timetable has already been pushed back several times: at the end of August, the US news agency had announced that the listing filing would be published within days. The company has never confirmed a date.

Investors value the company at between $1800 billion and $2000 billion, double its valuation in May ($965 billion). The offering could exceed the record set by SpaceX, which raised $85.7 billion in June.

In a preliminary version of the prospectus cited by the FT, Anthropic also reiterated to its potential investors that the AI it develops could pose “existential risks to humanity”.

Key takeaways

The established facts are limited: Anthropic presented figures to investors, and two media outlets reported on them. The rest, adjusted profitability, positive cash flow, return on computing power, falls under the company’s own statements, pending the publication of the prospectus. For Swiss readers following the generative AI revenue race, this document will be the first opportunity to compare these announcements with verifiable accounts.

Sources

  • Allnews: “Anthropic claims $16.8 billion in quarterly revenue”