Share with your CTO
The vision for AI-driven drug discovery hitting a structural ceiling, and the ceiling is data plumbing, not model capability. Cytiva’s head of applied AI, Jonathan Belcher, argues in MIT Technology Review that fully autonomous “dark labs” running continuous predict-test-optimize cycles are achievable within a few years, but only if wet lab instruments stop being closed silos. No AI-discovered drug has cleared FDA approval yet, though Belcher expects that within two to three years, contingent on labs generating FAIR data at scale rather than isolated instrument outputs.
What this means for your business
If your organization runs any computational biology, genomics, or drug development workload, the bottleneck Belcher describes will feel familiar before you finish the first paragraph. The AI models are ahead of the data infrastructure feeding them. Labs with proprietary instruments that don’t expose clean data interfaces are, in practice, running expensive equipment that can’t participate in an AI feedback loop, regardless of how sophisticated the models sitting upstream have become. That gap is the actual competitive disadvantage right now, not model choice.
Belcher’s argument holds, but it’s worth registering that Cytiva sells lab instruments and integration platforms into exactly the future he’s describing, which gives his optimistic two-to-three-year timeline for FDA approval of an AI-discovered drug a commercial incentive to land early rather than sober. The harder structural problem he surfaces, that most lab instruments are closed ecosystems, is credible precisely because it indicts the entire industry Cytiva competes in, including Cytiva’s own legacy install base. The FAIR data standard (findable, accessible, interoperable, reusable) isn’t a vendor invention; it’s a widely adopted research framework, and the gap between aspiring to it and achieving it at lab scale is where most enterprise life sciences IT budgets are quietly bleeding right now.
The decision this reframes isn’t whether to invest in AI models for drug discovery. It’s whether your instrument procurement contracts require open data APIs as a baseline condition. Every closed-ecosystem instrument added to a lab today is a future integration debt that compounds against whatever AI stack you’re building around it. If your next hardware refresh cycle doesn’t include interoperability requirements at the contract level, you’re not behind on AI strategy, you’re behind on the plumbing that makes AI strategy possible.
Concept deep-dive: FAIR data
FAIR stands for findable, accessible, interoperable, and reusable, a set of principles for scientific data management adopted widely in research and increasingly demanded by enterprise life sciences organizations. Think of it as the difference between a filing cabinet only one person knows how to use and a shared database with a consistent schema. The business connection is direct: AI models trained on FAIR data can generalize and improve across experiments, while models trained on siloed, inconsistently formatted instrument outputs plateau quickly regardless of their architecture.
Based on reporting from Closing the data loop in AI-driven drug discovery, originally published 2026-07-27 07:40:00.
