AI's Biggest Payoff May Be in Medicine, Not Just Data Centers
As AI infrastructure expands rapidly, experts argue that drug discovery could deliver the most significant returns, with AI helping to reduce unproductive experiments and improve clinical success rates.
The rapid buildout of artificial intelligence infrastructure — chips, data centers, cloud computing, and energy capacity — has dominated the technology landscape. But the most valuable application of that capacity may not be in digital products or automation. It could be in medicine, where AI is increasingly being used to accelerate drug discovery and improve the odds of clinical success.
At the World Economic Forum's 2026 Annual Meeting in Davos, Nvidia CEO Jensen Huang described AI as a five-layer stack: energy, chips, cloud infrastructure, models, and applications. The first four layers create capacity; the application layer turns that capacity into measurable outcomes. Few fields offer a clearer path from computational progress to economic value and human benefit than healthcare.
U.S. healthcare spending reached $5.3 trillion in 2024, or 18% of GDP, including $467 billion in prescription drugs. While only a fraction of that total goes directly to drug discovery, the figures illustrate the enormous resources already devoted to treating disease. A medicine that materially changes the course of a common or serious illness can create value in multiple directions: improving patients' lives, reducing the need for other care, helping people remain productive, and generating substantial returns for developers and investors.
The commercial scale of such breakthroughs is evident in the recent growth of tirzepatide. In 2025, Eli Lilly reported approximately $23 billion in sales for Mounjaro and $13.5 billion for Zepbound, bringing annual sales of the two brands based on the same molecule to roughly $36.5 billion. Those figures demonstrate the commercial potential of a medicine that transforms treatment of a widespread condition. The consequential question is whether AI can make such successes more frequent by reducing unproductive experiments and improving decisions along the way.
Drug discovery consists of a long sequence of decisions under uncertainty: which chemical scaffold or biological mechanism to pursue, which experiment to run next, which safety signals matter, and which patients are most likely to benefit. AI can help researchers rank possibilities, identify relationships across large and varied datasets, and design experiments that produce more useful information. It does not eliminate weak biology or poor data, but it can make the process more efficient when connected to high-quality experimental data and laboratories capable of testing predictions quickly.
The most effective AI systems create a learning loop in which each experiment sharpens the next decision. The quality of that loop matters far more than the sheer number of hypotheses a model can generate. AI can also enable entirely new ways of thinking about where new medicines might come from. For example, AI models for mass spectrometry have allowed scientists to understand an organism's chemical code more comprehensively, leading to the discovery of a new hormone that might capture the benefits of exercise and its development into a candidate medicine with positive Phase 1 trial results in just four years.
Yet the field remains nascent. In an August 2026 Nature Reviews Drug Discovery piece, leading authors assessed progress over the past decade and concluded that, despite extensive model development and benchmarking, evidence of clinically relevant impact remains disappointingly limited. They recommended a shift toward evaluating whether AI improves real drug-discovery decisions. That challenge should be welcomed, as it gives AI-enabled drug discovery the right scoreboard: translation.
Drug development has a stubborn failure mode: results that look compelling in a model or laboratory often do not hold up in people. Generating more molecules and hypotheses at greater speed does little to solve that problem if the resulting candidates simply enter the same attrition funnel. For executives and capital allocators, the central question is whether AI improves the odds of eventual clinical success at each stage, from early discovery through clinical trials. That requires evidence that models enable better choices of biology or chemistry, that candidates with a higher probability of success advance to the clinic, and that those candidates go on to produce stronger trial results and better patient outcomes.
AI can help scientists search chemistry and biology more intelligently, learn faster from every experiment, and direct human judgment toward the questions that matter most. Few outcomes would better justify the scale of today's AI investment than medicine, because breakthroughs that arrive more often and reach patients sooner give millions of people more healthy years of life, hope, and meaning.
6
