Artificial intelligence (AI) for science has rapidly gained momentum at an extreme pace. Models trained on the scientific record are generating hypotheses and even running experiments in automated labs. Within months, they have delivered new antibodies, catalysts, and materials that once took years.
That acceleration is real, and it matters. But here’s what should keep us up at night: the United States is not in first place. China is building biological infrastructure at a scale and speed that makes our efforts look quaint. If we do not step up now, we’re about to watch other countries own the future of medicine.
The question is what happens next.
Today, most of these systems are still learning from papers and proxies. That is like training airline pilots only in simulators. You can learn procedures and practice scenarios. But real capability comes from flying in conditions that are messy and variable, like fog, crosswinds, and turbulence. Without those perturbations, the system never learns how to adapt when it counts.
At Amazon Web Services (AWS), I saw the same principle in computation. Intelligence did not leap forward because algorithms suddenly became smarter. It leapt forward because data centers created an environment where models could run at scale, generate outcomes, and continuously improve. Rows of standardized machines, built for parallelism and reliability, turned experimentation into a system. Without that infrastructure, algorithms were just code on a whiteboard.
Biology needs the same kind of substrate. Without it, we are still guessing. With it, discovery starts to look predictable by design.
Drug development still leans on animal models and small patient cohorts to make billion-dollar bets. Those proxies teach us something, but they do not teach how a molecule behaves across the complexity of human biology. That is why nine out of ten drugs that succeed in animals fail in human clinical trials.
Biology needs an environment that gives intelligence the same systematic feedback that data centers gave to computation. That is what biological data centers provide. Robotic systems that sustain tens of thousands of standardized human tissues at once. Tissues that are vascularized and immune competent, clinically indistinguishable from patient biopsies under blinded review. Tissues that can be dosed, that bleed, that heal.
With that substrate, intelligence can move beyond proxies. Molecules generated upstream can be stress tested downstream in human systems. Doses can be chosen based on human-relevant endpoints. Toxicities can be caught before an investigational new drug (IND) application is filed. Diversity is built in, with donor-derived tissues reflecting women of childbearing age, pediatric populations, and rare subtypes.
Clinical trials remain. But they become confirmatory rather than exploratory. Instead of being cliff edges where risk concentrates, trials become predictable by design. I spent 15 years at AWS working in finance, bio, and scaling operations. The pattern is always the same. You build the infrastructure first, find the root cause, and systemize before you scale. That’s what makes this moment different. We finally have the operational foundation to turn drug discovery into a disciplined, repeatable process.
Policy is already shifting. The U.S. Food and Drug Administration’s modernization act and workshops on New Approach Methodologies show openness to human-relevant evidence as a complement or replacement for animal models. We’re on the NAM steering committee. We have a seat at the table.
Industry is under pressure. Even the most advanced AI models falter without high-quality input data. Patients, regulators, and investors are asking the same question in different forms. If the goal is human impact, why not start with human evidence?
Ethics and economics are converging. Patients are less willing to accept a system where animal success is treated as predictive. Investors are less willing to fund late-stage attrition. Better evidence earlier is both humane and efficient.
At AWS, I learned that infrastructure changes the possible—it’s the foundational anchor. Data centers did not make algorithms clever. They made intelligence inevitable by giving it a substrate to run, adapt, and scale.
Biology is now at the same threshold. Computational models can generate candidates at speed. But without standardized, scalable human evidence, they are still running in simulators. The countries that build this infrastructure first won’t just lead in drug discovery; they will define what medicine becomes for the next century.
That is why biological data centers matter. Robotic facilities sustain tens of thousands of human tissues. Tissues that can be dosed, that bleed, that heal.
When the evidence bleeds, discovery stops being guesswork. It becomes predictable by design. And if we want that design to bear an American signature, the time to act is now.
Julie O’Shaughnessy ([email protected]) is the Chief Operating Officer at Vivodyne. She was previously head of Global Operations and Strategy, Startup & Venture Capital Business Development at AWS.