Anthropic has filed its S-1 with the Securities and Exchange Commission, seeking to list on the Nasdaq with a valuation between $1.5 trillion and $2 trillion.
Joey Brookhart, an analyst at SemiAnalysis, joined Decoder to break down the filing. The document runs to hundreds of pages and details the economics of a frontier AI lab. Brookhart noted the company scaled efficiently to a $70 billion-plus annualised recurring revenue (ARR) base in a short timeframe. Gross margins, compute margins, and free cash flow figures met expectations, driving a rally in semiconductor and neocloud stocks.
Tokenomics and revenue sources
Brookhart runs SemiAnalysis’ Tokenomics model, which examines the unit economics of AI businesses. He pointed out that while Dario Amodei sees a path to $1 trillion in revenue, the source of that money remains unclear. The current market is narrow, with the top 1 percent of customers spending $50,000 to $100,000 per employee on AI tools. Broader enterprises are not spending heavily at this stage.
Spending at Anthropic is largely coding-specific, making it a single-vertical company. To reach a $1 trillion valuation, the business must expand into white-collar work where AI moves from augmentation to labour replacement. Brookhart cited Microsoft’s recent buyout offers to employees as an example of this shift, but noted there are currently no other verticals ready to move the needle.
Competition and market share
The industry faces external pressure from multiple fronts. Several labs, including OpenAI, DeepMind, and Meta Superintelligence Labs, possess similar training compute capacity. SpaceX plans to add ten net new gigawatts of power next year, with half dedicated to internal use cases.
In the first half of the year, Anthropic captured 90 percent of the industry’s net new ARR. By the third quarter, OpenAI had taken over as the majority driver of new ARR. Open-source models have grown from $1 billion in token inference costs in the first quarter to over $10 billion in the third quarter.
Brookhart highlighted that coding is unlikely to power a tripling of revenue from current levels. Future growth depends on verticalised use cases that go beyond simple automation.
What it means
The filing confirms Anthropic’s financial scale but does not resolve the question of how it will sustain that growth against rising competition. Enterprises are increasingly nervous about sending data to public cloud models, driving some law firms to run AI on-premises. Meanwhile, many companies are training models on open weights and running inference internally or externally. Brookhart noted that OpenAI’s lack of zero data retention in Fable slowed enterprise adoption, a problem the company plans to fix in the next release.
