Anthropic turned itself into a drug company in five months — and never showed the pricing

Mid-September, customers paid more than $1.70 for a token of Claude. By October the realized price — what customers actually paid — had halved, to $0.83, even as the number of tokens sold kept climbing ^[1]. On October 31 Anthropic filed papers asking the public market to value the same company in a different unit: not per token, but per cure. The S-1 calls life sciences "critical to our business," its fastest-growing sector, and sizes it against a $2 trillion global pharmaceutical market ^[2]. The pivot is real in its adoption. The calendar is where the argument lives. May 20: Bristol Myers Squibb deploys Claude across its global operations ^[3]. June 22: the confidential IPO filing, on annualized revenue above $30 billion ^[4]. June 30: a workbench for research scientists ^[2]. September 16: Novo Nordisk signs ^[5]. Then the $400 million stock purchase of the biotech Coefficient Bio and a board seat for Novartis chief executive Vas Narasimhan ^[2]. Two days before the filing, the founders pledged 80% of their equity to charity, with a provision letting Amodei keep voting control of the pledged shares through donor-advised funds ^[6]. None of this clock is Anthropic's alone: OpenAI bought Orchestra Bio eleven days before the S-1 ^[7]. Dario Amodei's month moves on the same calendar. On October 2 he said where he stood.

I've always thought of it as a marketing term. — Dario Amodei

On October 15 the register changed.

We must slow the pace at which we improve the capabilities of AI models. — Dario Amodei

By the October 31 filing it had moved again.

Biology is probably the area where scientific progress has the greatest potential to directly and unambiguously improve the quality of human life — Dario Amodei

No one at Anthropic has explained why the framing shifted. What got built in between is a matter of dates and contracts. The adoption is not nothing. Eighty-five percent of Eli Lilly's 50,000 enterprise-account employees actively use Claude, and Bristol Myers Squibb's stock rose 0.54% the day it announced the deployment. Novo Nordisk's chief executive cites productivity gains, and Anthropic reports two straight quarters of adjusted profitability ^[2][3][5][8]. Set that beside the other column of facts. Corporate AI spending stalled even as volumes rose, and two clients generated nearly a quarter of Anthropic's 2025 revenue ^[1]. And the prospectus spells out what a slowdown costs.

If our actual spend falls short, we must pay Google the difference. — Anthropic

Novo Nordisk, for its part, holds AI deals with OpenAI and Amazon at the same time as the Anthropic partnership — pharma is collecting labs, not committing to one ^[5]. The flagship biology claim is itself disputed: a University of Copenhagen computational biologist says he found the same enzyme systems years earlier and had shared unpublished work with Claude ^[9]. And clinicians closest to the work keep saying biology cannot be computed, only measured, and that no model solves human disease without a human-run lab ^[10]. The same prospectus that prices cures also warns the models could pose "catastrophic or existential risks to humanity" ^[11]. What the filing doesn't show is the price. The pivot is staffed, contracted, and board-certified on one side of the ledger. On the other, where the pricing and the margins should be, there is silence — no figure for what pharma pays per token, nothing for what the biology business earns. That one disclosure, pharma-specific revenue or pricing, is what November's buyers are still waiting on.