Drug Research and Development · global
Novo Nordisk Pursues Two Paths in AI Drug Discovery, From Biological Reasoning to Oral Macrocycles
One collaboration brings AI into research workflows, while another puts molecular design to the test through synthesis and screening. Both paths point to the same question: how can computational advances become drugs that can actually be used?
A series of experiments separates having AI propose research directions from turning the molecules it designs into oral drugs. In September 2026, Novo Nordisk announced two collaborations in succession: one with Anthropic to explore biological reasoning and R&D workflows, and another with Orbis Medicines to develop oral macrocyclic therapies for cardiometabolic diseases. The two address research judgment and molecular manufacturing, respectively, while laying out the validation hurdles that AI drug discovery must overcome.
The Anthropic collaboration, announced on September 16, will start with specific questions raised by Novo Nordisk scientists and computational teams. Initially, it will test Claude Science in selected R&D workflows and use Anthropic models to support software development. However, the announcement did not specify the datasets or scoring criteria used in testing, or how much time could be saved compared with existing methods. Accelerating drug development remains a goal of the collaboration, rather than a quantified outcome.
This also makes data governance and human oversight important conditions for the collaboration. Both parties said they had incorporated these into the design, but did not disclose specific implementation details. Whether the models' inferences about biological mechanisms are correct, and whether they can remain reliable across different research settings, still requires confirmation through reproducible evaluations and experiments. Fluent answers alone cannot establish effectiveness.
The Orbis collaboration, announced the following day, focuses on molecules that can actually be synthesized. Orbis's nGen platform combines generative AI, high-throughput synthesis and experimental screening, using proprietary macrocycle data continuously generated through experiments to iteratively refine candidates. Its goal is to improve the oral bioavailability of these molecules so that compounds with therapeutic potential can be administered orally.
According to Orbis's announcement, the collaboration covers multiple cardiometabolic disease targets, but their specific names were not disclosed. Orbis could receive up to a total of US$1.4 billion in upfront and potential development and commercial milestone payments, plus tiered sales royalties. Novo Nordisk will also make a strategic investment. This ceiling includes conditional future payments and is not an amount already paid. The announcement also provided no human efficacy or safety data for drug candidates under this collaboration, leaving the promise of oral therapies to be validated.
In the same week, Sanofi's choice illustrated another approach to resource allocation. On September 14, the company announced plans to transfer 20 established medicines and three manufacturing plants to Cheplapharm in exchange for a 26.4% equity stake, allowing it to focus more closely on innovative R&D. The transaction remains pending: under the plans at the time, the commercial transfer of the medicines would begin in the first quarter of 2027, subject to employee consultation, regulatory approvals and other closing procedures.
These arrangements show that large pharmaceutical companies are using external partnerships to gain access to research tools, chemistry platforms and different operational capabilities. But investment amounts and the scale of collaborations indicate only how many resources companies are willing to commit. For AI drug discovery, a more convincing next step is to disclose research performance that can be evaluated and have candidate molecules successively pass tests of drug properties, human exposure, efficacy and safety.