NVIDIA and Eli Lilly are putting $1 billion behind a simple bet: that artificial intelligence can make drug discovery move at something closer to internet speed than to the pace of traditional pharma.
The companies on Monday outlined a joint five-year investment to build an AI-driven drug discovery lab in the San Francisco Bay Area, with operations expected to start by late March 2026. The aim is ambitious: use generative models and tightly coupled automation to cut some early discovery timelines from years to months.
“This is about exploring vast biological and chemical spaces in silico before a single molecule is made,” NVIDIA chief executive Jensen Huang told attendees at the J.P. Morgan Healthcare Conference in San Francisco, where the deal was unveiled on January 12, 2026.
Huang cast the partnership as a shift in how medicines are invented in the first place, not just how they are tested.
Lilly’s chief executive, David A. Ricks, put it more bluntly: “The way we’ve discovered medicines for 150 years is too slow and too expensive. Alone, we can’t change that fast enough. Together, we think we can.”
A hybrid lab where coders and chemists share the same bench
The new co-innovation facility will sit in the Bay Area’s dense life-sciences corridor and is deliberately being built as a physical and cultural hybrid of Silicon Valley and Big Pharma.
Instead of shipping models and data back and forth across corporate firewalls, NVIDIA and Lilly plan to colocate their teams. NVIDIA will send a dedicated healthcare and life-sciences group led by Kimberly Powell, the company’s vice president in charge of that business. Lilly will embed biologists, chemists and clinicians, under the direction of chief information and digital officer Diogo Rau.
Powell described the plan as “lab-in-a-loop scientific workflows,” where AI models, robotic systems and human experts form a continuous feedback cycle.
“You run an experiment, that data feeds the model, the model proposes the next experiment—24/7, without waiting for the next project meeting,” she said in an interview. “We want cross-pollination happening in real time, not in quarterly steering committees.”
On one side of the facility, NVIDIA accelerated computing systems running the company’s BioNeMo platform—now built on its latest Vera Rubin architecture—will be stacked in rows. A short walk away, wet labs will be fitted with automated instruments, flow cytometers and robotic arms.
Digital twins of those labs, built in NVIDIA’s Omniverse and trained with its Isaac robotics tools, will let teams rehearse and debug complex workflows in simulation before they ever touch real equipment. In theory, that should cut down on downtime, failed runs and costly mistakes.
Trying to turn discovery into a closed-loop system
The basic problem the two companies are trying to attack is well known. Drug discovery takes too long and costs too much. Industry estimates put the average cost of bringing a new drug to market above $2 billion when you include the failures, often over more than a decade of work.
NVIDIA and Lilly say their lab is structured to hit that problem from several angles at once:
- Use generative AI models, trained on decades of Lilly’s proprietary data along with public datasets, to design and score potential drug molecules entirely on computers.
- Run automated experiments on the most promising candidates in high-throughput wet labs, with AI agents watching for failures and adjusting conditions on the fly.
- Feed the data from every round of experiments back into the models, so the system gets a little smarter and more tailored with each cycle.
The goal is to screen far more hypotheses than a traditional team could handle, while spending less time and money on dead ends. Flow cytometers, for example, will be paired with software that can spot abnormal patterns and recalibrate without waiting for a human technician to intervene.
Rau calls it “rapid experimentation with increasingly customized models,” a way to push thousands of hypotheses through the system in the time a conventional group might test a handful.
“We’re not promising to magically erase biology’s complexity,” he said. “But we are trying to compress the cycle of idea, test, learn, repeat. If that cycle shrinks from months to days, the whole pipeline moves faster.”
For now, the partners are not saying which specific stages of R&D—target identification, hit discovery, lead optimization or preclinical testing—they expect to speed up the most. Nor have they laid out how they will objectively measure those gains in a way investors, regulators and peers can compare against current practice.
Owning more of the AI stack
The $1 billion pledge is about more than bricks, robots and servers. It reflects a deeper shift in how both companies think about their competitive edge.
For NVIDIA, whose chips already dominate the AI accelerator market, the lab cements life sciences as a core strategic vertical, not just another customer category. The company is pushing BioNeMo from a research environment into what Powell calls a “full open development platform” for biology and chemistry, aimed at drugmakers, biotech startups and robotics companies that want to plug in.
An early ecosystem is forming around that stack. Multiply Labs, which uses robots to manufacture cell therapies, says it has cut costs per dose by about 70 percent and boosted throughput 100-fold after building on NVIDIA’s digital twin tools, according to its own figures. Other firms, including Basecamp Research and TetraScience, are layering their own software and services on top of NVIDIA’s platforms.
Lilly, meanwhile, is building on an “AI factory” initiative it announced in October 2025 and touted at the time as the most powerful pharmaceutical supercomputer in the business. The joint lab adds raw compute, specialized talent and physical lab capacity to that earlier program.
“Pharma used to outsource this kind of experimentation to vendors and point solutions,” said one industry consultant who advises both drugmakers and cloud providers. “Lilly is signaling that foundational AI and the data it runs on are now core infrastructure, like manufacturing plants.”
Winners, losers and skeptics
The project will likely sharpen existing tensions across both tech and healthcare.
Rival chipmakers such as AMD and a wave of specialized AI hardware startups are now under more pressure to prove they can match not just NVIDIA’s performance, but also its deep integration into real-world drug pipelines. Cloud providers that focus on renting out generic compute may feel squeezed if more pharma companies follow Lilly and opt to build vertically integrated “AI factories” under their own roofs.
In biotech, the picture is murkier. AI-first startups that once sold themselves as indispensable discovery partners to Big Pharma now have to contend with the prospect that large drugmakers will go directly to infrastructure providers like NVIDIA and keep more of the capability—and the value—in-house.
“Investors will start asking hard questions,” said a venture capitalist focused on AI–biotech deals. “Do you really have proprietary data or IP that Lilly plus NVIDIA can’t replicate at scale?”
So far, public criticism has been muted. There has been relatively little pushback from regulators, academic researchers or patient advocates, even though those groups will bear much of the risk if AI-designed drugs behave in unexpected ways, and reap the benefits if the bet pays off.
“Everyone is cheering the speed,” said a bioethicist at a major U.S. university, who asked not to be named because she consults for several pharmaceutical companies. “I’d like to hear a lot more about safety, transparency and who gets access to these therapies if they work.”
Regulation, risk and the black box problem
On those fronts, NVIDIA and Lilly are offering broad assurances, but few specifics.
Executives say the partnership will comply with U.S. privacy laws such as HIPAA and international regimes like Europe’s GDPR, given the sensitivity of the clinical and genomic data expected to feed their models. They point to internal safeguards for bias control, model validation and security.
The harder questions, though, sit with regulators.
How will the Food and Drug Administration evaluate a candidate whose structure was proposed by a black-box generative model trained on proprietary datasets? What standards will apply to the foundation models that quietly shape which molecules ever make it to the clinic?
“Right now, the FDA has more guidance for AI in medical devices and diagnostics than for AI at the very front of drug discovery,” the bioethicist said. “Someone is going to be the test case. This lab is volunteering for that role, whether they say it out loud or not.”
The partners have not explained how they will separate the impact of advanced AI from the impact of simply spending more—on compute, people and automation. Nor have they committed to publishing success and failure rates for AI-generated candidates, a metric many researchers see as essential for cutting through hype.
Ricks acknowledged the uncertainty but argued that waiting for perfect clarity carries its own risks. “If we stand still because the rules aren’t fully written, patients lose,” he said. “We intend to move fast, and work with regulators to make sure we move safely.”
A $1 billion promise—and a lot to prove
Key details of the $1 billion commitment remain opaque. The companies have not broken out how much is earmarked for capital expenditures like servers, lab equipment and facilities versus operating costs such as salaries, maintenance and specific research programs.
They also have not said whether any of the money is tied to milestones, performance targets or go/no-go decisions. Analysts expect at least some of those details to surface in future financial filings if the project is deemed material for investors.
For now, the number functions as much as a signal as a spreadsheet entry. A leading AI chipmaker and one of the world’s most valuable drugmakers are both publicly saying that AI in drug discovery is no longer a side experiment. In their view, it is now table stakes.
The Bay Area lab, once it opens later this quarter, will be an early test of whether that belief is justified.
If NVIDIA and Lilly can show that AI actually shortens timelines, trims costs and produces safer, more effective medicines, the experiment could reset expectations across the industry.
If they cannot, the gleaming facility in San Francisco may come to represent something else: a $1 billion reminder of how stubborn biology can be, even in an era defined by silicon.
