Novartis

Novartis AG is a Swiss pharmaceutical company headquartered in Basel, whose American depositary receipts trade in New York (NYSE: NVS). Formed in 1996 through the merger of Ciba-Geigy and Sandoz, it now focuses on innovative medicines after spinning off its generics business, Sandoz, as a separate company in October 2023. Its best-known products include the heart-failure drug Entresto, the psoriasis and arthritis drug Cosentyx, the breast-cancer drug Kisqali and the radioligand therapy Pluvicto. It appears here not as an AI vendor but as a large pharmaceutical AI adopter: it does not sell an AI product, and the AI it uses sits inside its own research and development.

Its defining pattern is model-licensing. Rather than building a frontier model of its own, Novartis pays AI-native labs for access to theirs and points them at its own targets. In January 2024 it signed a multi-target small-molecule collaboration with Isomorphic Labs, the Alphabet company spun out of DeepMind, paying 37.5 million dollars upfront against up to 1.2 billion dollars in milestones for three undisclosed targets. In February 2025 the two companies expanded it by up to three more research programs on the same financial terms.

The same pattern repeats across drug types. In September 2024 Novartis signed with Generate Biomedicines for protein therapeutics, from short peptides to antibodies, paying 65 million dollars upfront, including a 15 million dollar equity purchase, with more than 1 billion dollars in potential milestones. In November 2024 it paid Schrödinger 150 million dollars upfront for a multi-target collaboration plus an expanded three-year software license; Schrödinger's release frames its platform as physics-based computational modeling rather than AI, so that deal is best read as computational chemistry. In July 2026 Chai Discovery announced a collaboration giving Novartis access to its latest models, including Chai-3, for antibody discovery across multiple targets, after more than a year of technical work together. Terms were not disclosed.

The in-house half is data rather than models. The company's data42 program pools its own research and clinical data; in March 2020 Novartis said it had brought more than 2,000 clinical studies and roughly 2 million patient-years of data onto the platform and tested about a dozen machine-learning models. It announced an AI collaboration with Microsoft in October 2019, and in February 2026 broke ground on a San Diego research center, due to open in 2029 with about 1,000 employees, that it says will have AI, data and computational capabilities embedded throughout. Separately, chief executive Vas Narasimhan joined the board of Anthropic in April 2026, appointed by its Long-Term Benefit Trust: a governance role, not a commercial agreement.

The honest framing: Novartis's value rests on its marketed medicines and late-stage pipeline, not on AI. AI is a lever on discovery speed and trial design, and each licensing deal is an option: the milestone totals are ceilings paid only if programs succeed, and none of these collaborations has yet been reported to produce an approved medicine. Among peers in the catalog, Gilead also procures specialists, while Novartis concentrates on licensing frontier discovery models; Eli Lilly productized its own models through TuneLab, Moderna built hundreds of models in-house, Merck & Co. standardized on one enterprise AI platform with Google Cloud, and Novo Nordisk runs a multi-vendor stack across OpenAI, AWS and Anthropic.