Novo Nordisk’s OpenAI deal reaches far beyond drug discovery

Novo Nordisk, the Danish pharmaceutical company behind the blockbuster diabetes and obesity drugs Ozempic and Wegovy, announced in April that it will deploy OpenAI technology across drug discovery, manufacturing, supply chain operations, commercial functions, and employee training. Pilot projects are underway, and the companies have said broader integration is planned through 2026.

The agreement arrives as Novo faces intensifying competition in obesity treatment and works to regain momentum after a sharp decline in its share price. OpenAI has spent the past two years building relationships across the pharmaceutical industry, signing agreements with companies including Moderna, Sanofi, and Thermo Fisher. Novo’s partnership pushes that strategy deeper into the business than many earlier deals.

Why Novo is making this bet now

Novo helped establish the obesity-drug category with Ozempic and Wegovy, but Lilly’s Zepbound and Mounjaro have emerged as major challengers, and both companies are now racing to develop oral treatments that could expand the market beyond injectable drugs. In August, Mike Doustdar took over as chief executive, succeeding Lars Fruergaard Jørgensen under pressure to defend Novo’s position in one of the industry’s most valuable markets.

The OpenAI partnership also follows earlier AI investments by Novo, including a collaboration with Nvidia to use the Gefion sovereign AI supercomputer for drug discovery.

The most ambitious piece is the least defined

Two days after announcing the Novo partnership, OpenAI introduced GPT-Rosalind, its first model built for biology rather than general chat or coding. GPT-Rosalind is a reasoning model that works through a problem in steps instead of returning a single fast answer. It targets the daily work of a research lab: pulling together published evidence, generating hypotheses, designing experiments, and analyzing data. A general chatbot guesses at how two proteins interact. A model tuned on biology is meant to reason its way to the answer.

Novo was named as an early GPT-Rosalind collaborator alongside companies including Amgen, Moderna, and Thermo Fisher. Access to the model is currently limited to vetted enterprise customers participating in OpenAI’s trusted-access program.

Drug discovery has been one of the pharmaceutical industry’s most active areas for AI investment. Companies including Recursion, Exscientia, and Insilico Medicine have spent years building AI-driven discovery platforms, while large drugmakers have signed partnerships aimed at improving research productivity.

Novo has not disclosed which research programs it plans to target with GPT-Rosalind. That leaves the most ambitious part of the OpenAI partnership largely undefined. While Novo has named the areas it will apply the technology to—manufacturing, supply chain, workforce training, and commercial operations—it has said little about how the biology model will be used inside its research organization.

Novo’s manufacturing and supply chain plans

Novo’s least visible use of the technology may end up being its most concrete. The company plans to apply OpenAI’s models to manufacturing, distribution, supply chain, and commercial operations. Those functions became a focal point for Novo as demand for semaglutide-based treatments repeatedly outpaced supply, creating shortages and limiting patient access in several markets.

The mechanism here is the API, the connection that wires an outside model straight into a company’s own software, rather than having employees type questions into a chat window. That is what lets a model feed demand forecasts or production schedules directly into the systems that run a factory.

Thermo Fisher outlined a similar approach in 2025, embedding OpenAI’s models into its drug-development and clinical-research workflows to shorten trial timelines and flag failing therapies earlier. 

What workforce training produces

Novo also plans to train its workforce and raise what both companies call AI literacy, the everyday skill of using these tools at work.

Moderna ran this play first. Using OpenAI’s models, it built an internal assistant that more than 80 percent of employees adopted, and staff created over 750 custom assistants in about two months, including a pilot that checks clinical trial dosing. That is what workforce AI training tends to produce: many small employee-built tools and a daily-use habit.

The initiative arrives after Novo announced workforce reductions affecting roughly 9,000 employees in 2025 under CEO Mike Doustdar. The company has not linked those cuts to its AI strategy, but the timing sets modernization and contraction side by side as Novo reshapes the business.

Who stays accountable for AI decisions

Novo has framed the partnership with the usual safeguards: data governance, human oversight, and what it calls responsible AI use. The harder questions sit just outside the announcement. Those commitments arrive months after the FDA, in a March warning letter, cited Novo for systemic failures in how it collected, evaluated, and reported safety problems with semaglutide, including serious adverse events its procedures had wrongly screened out.

The two are unrelated, but each turns on whether a company can trust the systems it relies on to flag problems and act on them inside a highly regulated industry. Drugmakers remain responsible for safety monitoring and regulatory compliance, and for the decisions they make on what AI systems hand them.

For all the breadth of the announcement, the piece of it most likely to change how Novo discovers drugs is also the least settled. GPT-Rosalind remains available only to vetted enterprises through OpenAI’s trusted-access program, and Novo has not named a single research program or disease area it will point the model at.