Anthropic's Amended S-1 Shows $17B Q3 Revenue, Flags RSI Risk

Swing to a $567 million operating loss… compute and other commitments of about $895.4 billion

Anthropic’s amended registration statement (Form S-1/A) was posted on the website of the U.S. Securities and Exchange Commission (SEC) on Nov. 2 (local time). Revenue for the third quarter ended Sept. 30 was about $17 billion, with an operating loss of $567 million.

Third-quarter revenue rose from $11.6 billion in the second quarter and $1.4 billion a year earlier. Operating expenses were $17.5 billion, up from $10.6 billion in the second quarter, and operating income, which stood at $974 million in the second quarter, turned into an operating loss in the third quarter. The main cause of the loss was short-term compute (AI computing resources) costs; according to the filing, short-term compute is reportedly far more expensive than multiyear contracts. The filing also states that price cuts reduced revenue by about $600 million between the second and third quarters.

The net loss was $23.3 billion, mostly an accounting charge stemming from the increase in the value of convertible notes. Compute and other commitments total about $895.4 billion, and the company said it remains profitable on an adjusted basis that excludes stock-based compensation and charitable expenses (adjusted operating margin of 15%, versus 19% in the second quarter).

The amended prospectus also includes plans for long-term infrastructure investment on an unprecedented scale. About $895.4 billion in spending is planned, mostly related to computing capacity and data center partnerships with cloud hyperscalers (large cloud service providers). Of this, about $447.5 billion is to be spent by 2030, with about $69.8 billion to be deployed in the next 12 months alone.

Adoption by enterprise customers has become a key growth driver. The net revenue retention rate stands at about 900%, and the company has 250 enterprise customers spending more than $10 million a year. Meanwhile, on an adjusted basis, it recorded revenue of $33.24 billion and adjusted operating income of $4.08 billion over nine months. This suggests that the cost-per-task metric is improving as new algorithms are introduced, and that unit economics are improving as a result.

Also a warning that it is on the verge of recursive self-improvement (RSI) models

Meanwhile, the document also contains a warning that the company is on the verge of developing a model capable of recursive self-improvement (RSI).

RSI refers to the point at which AI becomes so good at training itself that AI capabilities accelerate dramatically. For those who view AI pessimistically, RSI symbolizes an irreversible catastrophe: AI could develop itself into an ever more powerful entity, escape human control and become indifferent to the harm it causes in pursuing its own goals.

For those who are optimistic about AI, by contrast, RSI (or the series of processes by which AI improves itself) shows the technology’s potential. It means AI could become a tool that delivers unimaginable productivity and drives scientific discovery, among other things.

Goldman Sachs and Morgan Stanley are leading the offering along with 20 co-managers. The company plans to list on the Nasdaq Global Select Market under the ticker “ANT” ahead of the U.S. Thanksgiving holiday. According to Bloomberg, the investment banks value the company at between $1.4 trillion and $1.7 trillion. The listing would be the first entry of a pure-play frontier AI laboratory into the public equity market, and attention is on whether it will set a new record.

Source: Form S-1/A

This article was translated from the Korean edition of Science Monitor with AI assistance and reviewed by our editors.