What BMS and Schrödinger Reveal About Pricing Agentic AI

TL;DR

Bristol Myers Squibb signed a deal to deploy Schrödinger's agentic AI co-scientist Bunsen, announced August 5, nine days after early access began. Bunsen lets a medicinal chemist run Schrödinger's physics-based workflows through an agent, scoped narrowly to the validated platform. No deal value was disclosed, but Schrödinger licenses its software by computational throughput rather than seats. My gut feel is that consumption pricing will be the default for agentic software in pharma R&D within two years.

Bristol Myers Squibb just signed a deal to deploy Schrödinger's agentic AI co-scientist Bunsen across its research organization. "Bunsen" entered early access on July 27. The agreement came nine days later. 2026 really is the year of the agentic pharma collaboration.

Schrödinger announced it on August 5, alongside Q2 earnings. BMS has been a long-standing Schrödinger software customer and collaborator, including under a 2020 discovery collaboration reported at $55 million upfront and up to $2.7 billion in milestones. Bunsen turns that into a different kind of business. Enterprise software, pushed out to scientists well beyond the computational chemistry group.

What Bunsen actually does

Schrödinger's physics-based methods, free energy calculations above all, have long sat behind specialist users. Bunsen's job is to let a medicinal chemist plan, run, and interpret those workflows through an agent, with RetroSynth integrated for synthesis planning. Schrödinger scoped the agent narrowly on purpose. It is optimized to execute the company's validated computational platform rather than act as a general-purpose scientific agent. For a large pharma research organization, that constraint is part of the selling point.

No deal value was disclosed. But it is interesting to look at the deal details. Schrödinger licenses its software based on computational throughput rather than seats. I have a lot of discussion with the pharma SaaS companies I advise around the best business model for agentic AI, so I'm always curious to see how others handle this.

My gut feel: Some form of consumption pricing will be the default for agentic software in pharma R&D within two years. Schrödinger simply got there first, because a physics platform was always metered by compute.