Anthropic S-1 Shocker, Greatest Business Ever? Google’s $500B Wipeout, Midterm Predictions
Anthropic's S-1 reveals an extraordinarily efficient, fast-growing business, but the hosts argue the AI buildout's power needs, treasury crowding-out and politics create risks, while Google's search slump looks overdone.
Watch on YouTube ↗Charlie's summary
- ▶ 5:41
Anthropic's $80 billion loss is a phantom number, and the business is actually capital efficient with rising margins
Sacks says the loss came from convertible notes issued to infrastructure partners like Amazon, which were marked up as Anthropic's valuation rose. It was not a cash loss. Anthropic raised roughly $110 billion, spent about $9 or 10 billion, and has about $100 billion left. Gross margin is around 85%. Sacks reads this as refuting the bubble narrative.
- ▶ 7:44
Compute multipliers: tasks get 10x cheaper each year, so Anthropic cuts prices and still grows margin
Sacks describes Dario's framework of raw compute plus compute multipliers such as algorithmic gains, better data curation and better architectures. A given task gets roughly 10x cheaper year over year. Anthropic passes about half of that on to customers and keeps the rest as margin. Sacks thinks this undercuts the claims that open models are commoditizing Anthropic, and that recursive self-improvement could widen its lead.
- ▶ 11:17
Anthropic is only about 1% penetrated, and profit is made at every layer of the AI stack
Freeberg counts customers spending $100K or more: 200 at end of 2024, 1,500 at end of 2025, and 11,900 by September. By his estimate about a million businesses worldwide could spend that much at under 1% of their total spend. His value-stack math says each dollar Nvidia spends becomes about $18 of Anthropic revenue. He assumes, without data, that customers earn 2 to 10 times what they spend. He sees a Google-style “TAM is commerce” thesis.
- ▶ 18:30
Anthropic's risk factors show compute is the binding constraint, and the AI buildout is becoming a political fight
Chamath had his team analyze the S-1's risk language. Compute dominates, especially data centers, financing commitments and power. Regulation comes second, while customer traction and talent loss barely appear. He argues Democrats gain from slowing power and data centers before 2028. Sacks says the party's agenda is bans or a “DMV for models”, and notes that capex is roughly 3% of GDP and drives current growth.
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.