Inside Anthropic’s trillion-dollar AI gamble
Semianalysis analyst Joey Brookhart argues Anthropic's S1 shows a capital-efficient, high-margin business, but its growth depends on spreading beyond coding and on the AGI training chase not outrunning revenue.
Watch on YouTube ↗Charlie's summary
- ▶ 3:02
Anthropic's revenue is still narrow: a coding-heavy business hoping to spread into white collar work, cyber and bio
Joey says the two leading labs have about 120 billion of ARR, yet the market is narrow. Only the top 1% of customers spend heavily, most Fortune 500 and Global 2000 firms spend little, and Anthropic's spend is mostly coding. Getting toward a trillion needs cyber, bio and white collar work, where AI moves from augmenting labor to saving or replacing it.
- ▶ 7:08
Tokconomics: inference margins went from negative 94% to positive 85% through token throughput gains
Joey says Anthropic's inference gross margin went from negative 94% two years ago to positive 85%. Token throughput on Trainium chips is up 5 to 10x year over year. Revenue per inference megawatt is about 65 million, versus mid-teens a year earlier. Only about 30% of compute serves revenue. The rest goes to research and training the next model.
- ▶ 13:14
A frontier lab has to stay at the frontier, since 80% of recent revenue came from models under six months old
Asked why Anthropic can't stop training and just sell what it has, Joey cites the S1 figure that 80% of recent revenue comes from models released in the last six months. Adversarial fields like cyber need frontier models, and the demand “will always be there for frontier intelligence.” Email summary doesn't. Enterprises often limit non-coders to Sonnet.
- ▶ 20:26
The $80 billion loss is a non-cash charge caused by Anthropic's soaring valuation, not an operating loss
Joey explains that warrants and convertible notes issued earlier must be marked up as the valuation rises. That creates a non-cash charge that looks bad if you don't know accounting. He notes Anthropic has burned only about 10 billion of cash and holds 120 billion plus in cash and securities. He doesn't expect dilution to matter if the AGI bull case holds.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.