NVIDIA has become the foundational AI infrastructure layer for biotech, anchoring partnerships with Eli Lilly and Thermo Fisher while its BioNeMo platform powers specialized foundation models at Terray Therapeutics and Apheris.1
BioNeMo gives pharmaceutical companies pre-trained biological models they can fine-tune for specific drug targets. Terray Therapeutics and Apheris are among the early adopters building on top of it, compressing timelines that once required years of wet-lab iteration.
Traditional pharma is restructuring R&D around this new stack. Novo Nordisk closed its internal cell therapy unit and licensed its Parkinson's disease assets to AI-native Cellular Intelligence.1 The move reflects a strategic calculation: partnering with AI-specialized developers is faster and cheaper than building internal capability.
Novo Nordisk is doubling down on GLP-1 — its core metabolic franchise — while offloading disease areas where AI platforms carry more comparative advantage. This externalization model is gaining traction across the industry as platform biology matures.
The broader platform landscape is accelerating. Boltz, Owkin, Basecamp EDEN, Edison Kosmos, and Natera all launched or expanded AI drug discovery platforms recently.1 The cluster of launches signals a shift from experimental research tools to production-grade infrastructure capable of supporting clinical pipelines.
Investor sentiment tracks the momentum. Conviction in leading names like Novo Nordisk remains strong, with the overall outlook for AI-driven drug discovery rated bullish and improving.1
The architecture emerging across the industry is a three-layer stack: compute infrastructure (NVIDIA), biological foundation models (BioNeMo-based platforms), and application-layer biotech firms. Companies at each layer are locking in partnerships before the market consolidates.
Drug discovery has historically suffered from high attrition rates and decade-long timelines. AI platforms targeting target identification, molecule generation, and clinical trial design attack the problem across the full pipeline — not just a single bottleneck.
The competitive question now is which foundation model providers and platform builders achieve the critical mass of validated results needed to become default infrastructure — the equivalent of cloud compute for biology.