How Anthropic’s IPO Could Help Finance Its GPUs — Jamin Ball, Altimeter
Jamin Ball reads Anthropic's S1 as a highly profitable, compute-constrained frontier business, argues scary headline figures are accounting artifacts, and explains why debt markets, not equity, are the real bottleneck.
Watch on YouTube ↗Charlie's summary
- ▶ 17:16
Anthropic's S1 shows a 7x inference efficiency gain, making it “almost accidentally profitable”
Jamin notes compute spent on serving models rose 2.3x over a year while revenue rose 14x, which implies roughly a 7x efficiency gain. A megawatt that once earned $10-15 million now earns $70-100 million. The S1 shows 85% gross margins and a 21% free cash flow margin in the latest quarter. The company raised $141 billion and holds $134 billion in cash, so about $7 billion burned. Jamin also notes free cash flow is still 8% positive after adjusting for stock-based comp.
- ▶ 19:48
Expect Anthropic's profitability to flip back as it reinvests, because it probably underinvested in compute
Jamin guesses the positive free cash flow shows Anthropic can “flip the switch” to profit, and that it will now spend aggressively on compute. Margins may level off or go negative again. Only about 27% of its compute serves revenue, so more serving compute could raise revenue without new capacity. He says to read any margin drop as a conscious investment choice, not a weakening business.
- ▶ 23:21
Token spend is a tiny share of corporate budgets today, and Stripe's expected jump hints at how much room remains
Half of the Fortune 500 uses Anthropic, yet they spend only 0.1% of expenses on it. Jamin cites Stripe at about 1% of expenses on tokens and expecting 15-20% in a few years. If Fortune 500 firms went from 0.1% to 1%, that would be another order of magnitude of growth. The open question is how deep Anthropic can penetrate its market.
- ▶ 26:28
The whole bull case reduces to three questions, and 80% of revenue comes from models under six months old
Jamin says the debate comes down to whether Anthropic can stay on the frontier, whether a durable business exists there, and what it costs. In the three months to September, 80% of revenue came from models six months old or younger. He expects open-weight models to win on token volume but frontier models to take most of the dollars. The $856 billion of commitments should be read as roughly $100 billion a year, which revenue growth can plausibly cover.
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.