Large pharmaceutical companies say artificial intelligence, built on models like those from Anthropic, is shrinking the time it takes to discover new drugs. That is the customer-facing half of the story: AI as a tool that makes an expensive, slow process faster for the companies buying it.
The other half, less visible to drugmakers, is what that same AI costs the company supplying it. Anthropic's recent IPO raised money that is funding a wider boom in chip purchases, the infrastructure behind models like the ones pharma companies are using. But falling AI prices, the per-use cost of running those models, are straining the compute commitments Anthropic has already signed up for.
Put together, the two stories describe a single mechanism from both ends. Pharma companies benefit from AI getting cheaper and more capable; Anthropic has to fund the computing power behind that same improvement, and the economics of doing so get harder precisely because customers are paying less per use. Cheaper AI is good news for the buyer and a margin problem for the seller.
What is not yet clear is how much of Anthropic's compute spending is covered by long-term contracts, versus exposure that would reprice downward if AI prices keep falling, and whether that strain shows up in its reported numbers before or after further price cuts.