Novo Nordisk is expanding its use of artificial intelligence from workplace tools into the experimental loop of drug discovery. Under a new agreement, Amazon Web Services will become the pharmaceutical company’s preferred cloud provider and a strategic AI partner, with the two companies establishing a co-innovation hub at Novo Nordisk’s existing London facility.
AWS engineers will work alongside Novo scientists and data specialists. The ambition is to reduce handoffs between computational analysis and laboratory testing, particularly during the long path from identifying a biological target to administering a candidate medicine in a first human study.
A joint hub brings software closer to the laboratory
The partnership centers on agentic workflows: systems designed to select tools, coordinate tasks and pass results between stages of an analysis. Through Amazon Bio Discovery, researchers can access more than 40 biological AI models and combine AWS services with Novo Nordisk’s proprietary models.
Potential tasks include target identification, therapy design, analysis of biological information and the generation and ranking of candidate molecules. Promising candidates would still be synthesized and tested in a laboratory. Experimental results could then return to the computational workflow for another round of analysis and refinement.
That loop is more specific than asking a general-purpose model for an answer. It links models to scientific data, laboratory decisions and measurable results. It also leaves important details unresolved. Novo Nordisk has not described the precise functions individual agents will perform or how much authority they will receive at each stage.
AWS says its Forward Deployed Engineers will be embedded with Novo teams to build systems around practical research problems. Amazon Bedrock is expected to support work across clinical, genomic and imaging data, while AgentCore will provide tools for deploying and operating agents.
Existing deployments show speed, not independent discovery
The companies already have a substantial technology relationship. More than 25,000 Novo Nordisk employees have used a generative AI platform built on Amazon Bedrock, according to the announcement, supporting over 2,500 chatbot use cases. One deployment covers roughly 140,000 documents and handles more than 26,000 prompts each month.
Novo Nordisk has also used Anthropic’s Claude 3.5 through Bedrock for clinical documentation. An AWS case study says some document-generation tasks became more than 90 percent faster. Work that previously involved 40 to 50 people and could take up to 15 weeks was reduced to minutes for a team of three.
Those figures are vendor-reported and relate to documentation, not proof that AI can discover a successful drug at comparable speed. Medical professionals continue to review and validate the generated material. The distinction matters because a fluent document can be checked relatively quickly, while a biological hypothesis must survive laboratory experiments, safety assessment and clinical trials.
The partnership nevertheless provides a foundation for the new hub: broad employee adoption, centralized model access and experience placing human review around generated output.
The scientific bottleneck moves rather than disappears
Agentic AI may reduce the time researchers spend transferring data, configuring separate tools or manually narrowing large sets of candidates. It could also make iterative work more continuous, allowing experimental evidence to inform the next computational step without rebuilding the workflow each time.
But faster selection does not guarantee a better medicine. Models reflect the quality and scope of their training data, and biological systems can behave differently from computational predictions. Laboratory validation remains the point at which a plausible candidate confronts physical reality, while human studies introduce further questions about safety and effectiveness.
Novo Nordisk’s broader AI activity includes work with OpenAI and Denmark’s Gefion supercomputer. The AWS agreement adds another route, focused on cloud infrastructure and coordinated model workflows rather than a single exclusive model provider.
The most meaningful test will therefore be operational, not promotional. If the hub can shorten repetitive analytical steps while preserving traceability and expert review, it may give scientists more opportunities to test strong hypotheses. The agents are being placed deeper inside discovery, but the evidence required to turn their suggestions into medicines remains firmly outside the model.
