Novo Nordisk and AWS Form AI Partnership to Accelerate Drug Discovery

The partnership extends an existing relationship involving generative AI efforts. An AWS case study describes a Novo Nordisk system for clinical study reports that uses Anthropic's Claude 3.5 through Amazon Bedrock, alongside MongoDB and AWS services. AWS reports that the prior manual reporting workflow could take up to 15 weeks and require coordination among 40 to 50 professionals.

Clinical study reports are regulatory documents, so automation in this setting requires accuracy, consistency, traceability, and controls around sensitive data. Louise Skov, Novo Nordisk's head of content digitalization, told AWS: "We knew early on that this wasn't something we could solve internally. We needed expert knowledge from outside the pharmaceutical industry to move forward with confidence."

Insider Monkey reported that Novo Nordisk intends to connect genomic, imaging, and clinical data so that early research can inform clinical-trial design. That is a materially broader use case than document generation: it combines heterogeneous scientific and clinical data, where data quality, provenance, access control, and evaluation against existing research processes are central engineering constraints.

Insider Monkey noted that the companies did not disclose financial terms or quantified targets for research savings or development timelines. The available reporting therefore establishes the partnership's infrastructure and intended applications, but not quantified research savings or changes in development timelines.

Across comparable pharmaceutical AI programs, cloud platforms can speed data access and experimentation, but demonstrating drug-discovery value generally requires evidence at successive stages, from target validation through preclinical and clinical outcomes. For ML and data teams, the London hub is notable because it places platform engineers and applied AI specialists alongside domain researchers, an operating model often used when generic foundation-model tooling must be adapted to regulated scientific workflows.