TL;DR
Novo, recently renamed from Novo Nordisk, has signed Anthropic into its research operation, using Claude models and the newly launched Claude Science for scientific reasoning and agent-driven software work. It is the fourth major AI partnership the company has announced. Chief executive Mike Doustdar wants it to be the most AI-driven healthcare business anywhere.
What has actually shipped so far
One thing, and it is worth noting because it is not drug discovery. Novo built NovoScribe on Claude to produce clinical study reports, compressing a job measured in months into one measured in minutes. Regulatory documentation, in other words — expensive, highly structured, and the sort of work where a language model’s strengths are real.
The discovery claims are forward-looking. Anthropic and Novo say they will work through problems the drugmaker’s scientists and computational teams nominate, and build tooling for biological reasoning. Dario Amodei framed it as shortening research timelines and improving outcomes, which is the standard promise.
Four vendors, one research pipeline
Anthropic joins a queue. Novo runs a London AI hub with AWS using Amazon Bedrock and Bio Discovery to find targets and link genomic, imaging and clinical data. OpenAI is embedded across internal processes from research to commercial. NVIDIA is building bespoke generative models for early research and trials.
Running four frontier suppliers at once is a hedge, and an expensive one. It also sits awkwardly against Gartner’s recent advice that the frontier labs are not yet enterprise grade — guidance aimed precisely at organisations doing what Novo is doing.
Looking forward
The London hub is the part UK readers should track, since it puts frontier-model drug discovery work inside the UK rather than in Denmark or California. The context is less flattering: Doustdar is running a turnaround while Novo loses ground in diabetes and obesity, which is a familiar moment for large AI commitments. Watch whether anything beyond NovoScribe reaches production, because report generation working is not the same as discovery working.