Anthropic files for IPO: what the S-1 reveals about the economics of AI
AI company Anthropic has filed an S-1 document to list on the Nasdaq, giving the first audited look at the finances of a leading frontier AI lab. SemiAnalysis analyst Joey Brookhart calls the results strong, but says future growth sources remain unclear.
Anthropic, the US artificial intelligence (AI) company behind the Claude models, filed an S-1 this week — the registration document for a public share offering on the Nasdaq. According to The Verge, trading could begin as soon as next month at a valuation between $1.5 trillion and $2 trillion. The filing offers the first audited look at the economics of a leading frontier AI lab.
The analyst's view
The Verge's Decoder podcast discussed the document with Joey Brookhart, an analyst at the chip and AI research firm SemiAnalysis. His team built an outside model of Anthropic's finances back in July. Brookhart says the gross margin, compute margin and free cash flow figures in the filing are what investors would want to see. He cites more than $70 billion in annualized recurring revenue (ARR) and says semiconductor, neocloud and compute stocks rallied again in the two days after the filing.
According to him, Anthropic's gross margin on inference — running finished models for customers — moved from negative 94 percent two years ago to a positive 85 percent. Revenue per inference megawatt is about $65 million, versus the mid-teens a year earlier. Compute costs run a little over $10 million per megawatt. Only about 30 percent of compute goes to revenue-generating inference; the rest is used for research, experimentation and model training.
Open questions
The analyst notes that the market is still narrow: the two leading labs together have about $120 billion in ARR, and much of Anthropic's revenue comes from coding. Reaching $1 trillion in revenue would require a broader move into white-collar work, where AI shifts from augmenting staff to saving or replacing labor. Competition is growing as well: in the third quarter OpenAI probably captured most of the industry's net new ARR, and open-source inference spending rose from roughly $1 billion to more than $10 billion in ARR terms.
