Can AI Actually Discover Drugs? Novo Nordisk Is About to Find Out.
Can AI Actually Discover Drugs? Novo Nordisk Is About to Find Out.
Analysis | May 17, 2026
Earlier this year, Novo Nordisk struck a deal with OpenAI to deploy artificial intelligence across every function of the company — drug discovery, clinical trials, manufacturing, supply chain, and corporate operations. The announcement was greeted with the usual mixture of investor enthusiasm and scientific scepticism that meets most pharma-AI partnerships. But Novo’s situation is specific enough that the deal deserves a closer look than the press release cycle gave it.
The Danish company makes Ozempic and Wegovy. For three years those two drugs made Novo Nordisk the most valuable company in Europe by market capitalisation and reshaped the global conversation about obesity treatment. That window is narrowing. Eli Lilly has closed the gap, and in some markets overtaken it. The competitive arithmetic now requires Novo to find the next thing faster than it has found things before. That is the actual context for the OpenAI partnership — not transformation for its own sake, but a company with a clear and urgent deadline trying to use every available tool.
What AI Is Actually Being Asked to Do
The research application is where the scientific stakes are highest and the uncertainty is greatest. Novo wants to use OpenAI’s models to analyse complex biological datasets, identify promising drug candidates earlier, and compress the time between a promising compound and a clinical trial. Drug discovery has been held out as AI’s killer application for a decade. The results have been uneven. Early-stage promise has repeatedly run into the biological complexity of late-stage trials, and the graveyard of AI-derived compounds that failed in humans is already well populated.
What is different now is the capability of the models and the quality of the data they can be trained on. The biological datasets available in 2026 are substantially richer than those from five years ago, and the models themselves can handle the kind of multi-variable reasoning that earlier systems could not. Whether that is enough to actually change drug discovery timelines in a meaningful way is still an open question. But it is a more credible question than it was.
The supply chain application is less scientifically interesting and probably more immediately valuable. Wegovy and Ozempic have both faced supply shortages since launch — a problem that cost Novo real revenue and handed Lilly market share at a critical moment. AI-assisted manufacturing optimisation and demand forecasting will not generate headlines, but if it means the company can produce and distribute its drugs more reliably, the return on investment is direct and measurable.
Why Pharma Keeps Betting on AI
Novo is not alone in this. Eli Lilly, AstraZeneca, Pfizer, and Roche have all announced AI partnerships in the past 18 months. The wave has two drivers that are worth separating. The first is genuine scientific optimism: the models are better, the data is richer, and the computational cost of running experiments in silico rather than in a lab has dropped dramatically. The second is investor pressure: pharmaceutical companies that cannot point to an AI strategy are being penalised in their valuations regardless of whether the underlying science justifies it.
Novo’s partnership is distinguished by one specific commitment: full integration by the end of 2026. That timeline is aggressive, and the reality will almost certainly be a phased rollout that extends into 2027. But announcing a deadline publicly creates a form of accountability that looser partnership language does not. It is a signal that this is a strategic priority rather than a research experiment.
The Question the Next 18 Months Will Answer
The most important unknown is whether OpenAI’s general-purpose models are actually the right tool for drug discovery, versus specialised biotech AI platforms built specifically for protein structure prediction and molecular modelling. Companies like Isomorphic Labs, Recursion Pharmaceuticals, and Insilico Medicine have spent years building domain-specific systems for exactly this problem. Novo is betting that the breadth and reasoning capability of a frontier general model outweighs the depth of a specialised one.
That is not an obvious call. It is, however, a testable one. By the time Novo reports full-year 2027 results, there should be enough pipeline data to start answering whether the OpenAI partnership accelerated anything that matters. Until then, the announcement is the bet.
