Jensen Huang at Recursion: how AI is enabling leaders in drug discovery

01 What Recursion is doing in AI drug discovery

Recursion Pharmaceuticals has built one of the most ambitious AI drug discovery platforms in the industry. The company combines high-throughput phenomic screening — imaging cells under thousands of perturbations — with machine learning to identify novel drug candidates without relying on traditional target-based hypotheses. Its proprietary platform maps cellular biology at scale, generating petabytes of biological images that feed neural networks trained to recognize disease-relevant phenotypic signatures.

The approach is fundamentally different from conventional drug discovery. Instead of starting with a known protein target and searching for a molecule that binds it, Recursion starts with disease-state cell images and lets algorithms surface patterns that human researchers might miss. This phenotypic-first strategy has yielded a pipeline spanning oncology, rare diseases, and inflammation. Jensen Huang's appearance at Recursion signals NVIDIA's belief that this kind of industrial-scale biological data processing is a defining use case for accelerated computing.

02 How NVIDIA powers pharmaceutical AI

NVIDIA's role in AI drug discovery extends well beyond supplying GPUs. The company has developed BioNeMo, a generative AI platform for drug discovery that supports protein structure prediction, molecular docking, and generative chemistry. Through partnerships with companies like Recursion, Schrödinger, and Insilico Medicine, NVIDIA provides the computational infrastructure — from Clara computing platforms to cuQuantum simulation tools — that makes large-scale biomolecular modeling tractable.

The collaboration between NVIDIA and Recursion includes a supercomputer built specifically for drug discovery, reportedly among the most powerful in the pharmaceutical industry. This infrastructure enables Recursion to train foundation models on biological data at a scale that was previously impossible. NVIDIA's investment in this space reflects a broader thesis: that the next decade of pharmaceutical innovation will be computationally driven, and the companies with the best AI infrastructure will discover drugs faster, cheaper, and with higher success rates than those relying on wet-lab methods alone.

03 The shift from wet lab to computational

Traditional drug discovery has been described as one of the most expensive and time-consuming endeavors in science. The conventional pipeline — from target identification through hit finding, lead optimization, and preclinical studies — can take 9 to 12 years and cost billions of dollars, with failure rates exceeding 90%. The vast majority of compounds that enter clinical trials never reach patients. This inefficiency has driven the pharmaceutical industry to seek computational approaches that can compress timelines and improve success probabilities.

AI drug discovery promises to invert this model. By simulating molecular interactions, predicting protein structures, and generating novel candidate molecules computationally, AI can screen millions of compounds in hours rather than months. The shift does not eliminate wet-lab work — experimental validation remains essential — but it dramatically reduces the search space. Companies report compressing the hit-to-lead phase from years to months, and the total time from target identification to investigational new drug (IND) filing from a decade to under five years in some cases.

04 What AI drug discovery can do that traditional cannot

AI drug discovery excels at tasks that are computationally intractable for traditional approaches. Generative models can design novel molecular structures that do not exist in any chemical library, optimized for specific properties like binding affinity, solubility, and metabolic stability. Protein structure prediction tools, exemplified by AlphaFold, can determine three-dimensional protein structures from amino acid sequences in minutes — a task that previously required years of experimental crystallography.

Beyond molecular design, AI enables phenotypic screening at unprecedented scale. Recursion's platform images cells under thousands of genetic and chemical perturbations, then uses computer vision to identify subtle phenotypic changes that correlate with disease states. This approach can discover unexpected drug-disease relationships — repurposing existing drugs for new indications or identifying novel targets that hypothesis-driven research might overlook. The ability to integrate multi-modal data — genomics, proteomics, imaging, and clinical records — into unified models is something traditional drug discovery simply cannot do at scale.

05 The pipeline of AI-discovered drugs

The AI-discovered drug pipeline has grown rapidly. Recursion alone has multiple candidates in clinical trials, including REC-2282 for neurofibromatosis type 2 and REC-4881 for familial adenomatous polyposis. Insilico Medicine's INS018_055, an AI-designed drug for idiopathic pulmonary fibrosis, reached Phase 2 clinical trials — a milestone for AI-generated molecules. Exscientia, Schrödinger, and BenevolentAI each have candidates progressing through various stages of clinical development.

The number of AI-discovered molecules entering clinical trials has grown from near zero in 2020 to dozens by 2026, reflecting both the maturation of AI platforms and growing confidence from regulators. However, the pipeline remains early-stage: most AI-discovered candidates are in Phase 1 or Phase 2 trials, and no AI-designed drug has yet completed Phase 3 and received regulatory approval. The true test of AI drug discovery will come as these candidates advance through late-stage trials where efficacy in human patients — not computational elegance — determines success.

06 The regulatory and clinical trial implications

Regulatory agencies are adapting to AI drug discovery, but the framework remains evolving. The FDA has established pathways for AI-assisted drug development but requires that AI-generated candidates pass the same safety and efficacy standards as traditionally discovered drugs. The challenge for regulators is evaluating not just the drug candidate but the computational methods that produced it — raising questions about algorithmic transparency, training data quality, and reproducibility that traditional drug review processes were not designed to address.

Clinical trial design may also shift. AI can identify patient populations most likely to respond to a given drug by analyzing biomarker data, potentially enabling smaller, more targeted trials with higher success rates. Adaptive trial designs powered by AI could adjust enrollment criteria in real time. However, these innovations require regulatory buy-in, and the FDA has been cautious about approaches that could introduce bias or reduce statistical rigor. The interplay between AI innovation and regulatory caution will shape the pace at which AI-discovered drugs reach patients.

07 What this means for patients

For patients, the promise of AI drug discovery is ultimately about access to better treatments faster. If AI can compress drug development timelines and improve success rates, the cost of developing new drugs could fall substantially — potentially making treatments more affordable and expanding the range of diseases that are commercially viable to target. Rare diseases that have been neglected because traditional drug discovery economics did not justify the investment could become addressable as AI lowers the cost of discovery.

The Recursion-NVIDIA partnership exemplifies a broader trend: the convergence of computing and biology. Jensen Huang's advocacy for AI in drug discovery reflects NVIDIA's conviction that accelerated computing will transform healthcare the way it has transformed language processing and image recognition. For patients with conditions that have no effective treatments today, that transformation cannot come soon enough — though the gap between computational promise and clinical reality remains real, and the first AI-designed drugs must still prove themselves in human trials.