Roche puts artificial intelligence at the centre of its drug strategy

Roche is putting artificial intelligence at the centre of a pipeline that could deliver up to 20 new medicines by 2030. The Basel-based pharmaceutical group set out the strategy at its Pharma Day on September 28, 2026, presenting AI as a tool to increase the pace and precision of research and development.

Roche has spent the past several years rebuilding momentum after research setbacks and revenue pressure caused by expiring patents in 2022. Teresa Graham, head of the pharmaceuticals division, told investors that the company had returned to a stronger footing through a focused development programme.

The potential rewards are substantial. Roche said some candidates could become major products, including the breast cancer drug giredestrant, which has possible peak sales of more than CHF 30 billion. Those figures depend on successful clinical trials, regulatory approval and market uptake.

For Switzerland, the shift matters beyond Roche’s headquarters in Basel. The company is one of the country’s most important industrial research groups, with decisions taken in its laboratories and executive offices influencing jobs, investment and access to future treatments. AI now sits inside that long-term calculation, alongside biology, clinical evidence and commercial discipline.

Roche says it has advanced six programmes into the regulatory phase and moved three more into Phase II since its previous Pharma Day. Graham said the company had achieved its targets in 10 regulatory-relevant trials, giving investors a measure of progress behind the 2030 ambition.

The company has also tightened the way it decides which projects deserve further investment. Roche introduced a framework called “The Bar” at the end of 2023. Drug candidates must meet defined criteria before moving through the pipeline, a process designed to stop weak programmes early and concentrate resources on those with stronger prospects.

Levi Garraway, Roche’s chief medical officer, described the framework as the guiding principle for research and development. He said the company had reduced the number of projects while increasing the overall value of the remaining portfolio.

The reported results suggest that the strategy is changing the shape of Roche’s pipeline. Projects that reach Phase III, the stage involving large-scale testing before possible approval, had a 65% success rate in 2025. The rate has risen to more than 80% so far in 2026, according to Roche. These figures cover development performance, not guaranteed approvals or market launches.

Roche wants AI involved from the first biological question to the final optimisation of a drug candidate. The company’s plan goes beyond using software for isolated data analysis. It aims to integrate models throughout discovery, including disease understanding, the search for drug targets and the selection of compounds for further testing.

Aviv Regev, who leads Genentech’s Research and Early Development department, presented a “Lab-in-the-Loop” approach. AI models analyse large volumes of biological and chemical data, generate predictions and guide experiments. Scientists then test those predictions in the laboratory, feeding the new results back into the models.

That cycle could help researchers identify promising compounds earlier and reduce the number of unsuccessful experiments. It could also support more precise decisions about which candidates should enter costly development programmes. The impact will depend on the quality of the data, the design of the models and the ability of scientists to verify results.

Regev also set a clear boundary. AI can improve predictions and prioritise experiments, but it cannot remove the need for human clinical trials. Roche’s strategy therefore links computational systems with laboratory work, patient studies and regulatory review rather than treating AI as a substitute for them.

Roche’s pipeline push comes as the company manages the financial consequences of patent expiry and earlier research setbacks. Mature products can generate large revenues, but their protection eventually weakens, exposing pharmaceutical groups to competition and forcing them to replace sales with new treatments.

That pressure helps explain Roche’s focus on speed, selectivity and development costs. Garraway said the stricter “The Bar” process had reduced the number of projects while raising the value of the portfolio that remained. Roche also reported shorter development times and lower costs.

The commercial scale of the strategy is illustrated by giredestrant, an investigational breast cancer treatment with potential peak sales above CHF 30 billion. The figure signals the value Roche places on successful medicines in large disease areas, while also highlighting the risks attached to clinical development. A promising laboratory result can still fail in human trials, encounter safety concerns or face regulatory restrictions.

Roche’s approach combines a narrower project portfolio with a broader use of AI. That combination could help the company direct capital towards candidates with stronger evidence. It will also expose Roche to the limits of predictive models, particularly when biological systems behave differently from the patterns found in existing data.

Roche’s 2030 target will be judged by approved medicines and patient outcomes, not by the number of AI systems deployed. The company has set out a sizeable ambition: as many as 20 new active substances could reach the market within four years of the 2026 Pharma Day.

For Switzerland, delivery would reinforce Basel’s role as a global centre for pharmaceutical research and keep high-value development work anchored in the country. It could also bring new treatments to patients, including people with cancer and other serious diseases. Those benefits remain conditional on evidence gathered through clinical trials and decisions by regulators and health authorities.

The strategy will face practical tests. Roche must maintain reliable data flows between computational teams and laboratories, protect sensitive research information and ensure that AI-generated recommendations can be explained and reproduced. Researchers must also identify errors before they influence costly trials or patient care.

The company’s own figures provide an early benchmark. Roche reports stronger Phase III performance, shorter development times and a pipeline with several programmes nearing regulatory review. Over the next four years, investors and patients will see whether that operating model can convert predictions and experiments into approved medicines.