Alexander Wissner-Gross, a computer scientist and founder of Reified, gave a deadline for what he called the next field to be "bulk solved." Within "two to three years, maximum," he expects there to be small-molecule drug interventions, proposed in silico (on a computer rather than in a lab), that cure the top 5,000 diseases. He said they would come from a virtual cell: an AI model that predicts how a living cell responds to a drug or a genetic change.
Wissner-Gross made the forecast on Moonshots with Peter Diamandis, recorded on October 13, 2026 and published two days later. The episode had just discussed OpenAI's release of hundreds of AI-produced math results at once. Host Peter Diamandis, founder of XPRIZE and Singularity University, moved the conversation on: "Let's talk about bulk solving science, in particular biology."
What the $1.8 billion buys
Diamandis's starting point was an announcement from the Chan Zuckerberg Biohub on October 7. It is a $1.8 billion expansion of Biohub's virtual biology initiative, which he described as the largest coordinated effort yet to build an AI-ready biological dataset and a virtual cell. In his summary, Biohub put in $500 million, the Department of Energy $500 million through its Genesis Mission, the National Institutes of Health $500 million worth of datasets, and Google DeepMind and Isomorphic Labs $300 million.
Biohub's own announcement breaks the money down a little differently. Biohub's founding commitment is $500 million. The Department of Energy pledged more than $500 million over five years. The NIH share is not new cash: it consists of datasets and infrastructure built with more than $500 million of earlier federal investment. And Meta is named alongside Google DeepMind and Isomorphic Labs in the $300 million commitment. The money pays for biological measurements, from cryo-electron tomography to imaging of living tissue and experiments that perturb cells, along with shared standards so that data gathered separately can be used together to train models. The stated goal is open data supporting models that predict how cells respond to interventions.
Diamandis called this "the holy grail of biology" and described the ambition in personal terms: put your DNA into a model, simulate your cells, and learn which drug will work for you. He quoted Biohub's head of science, Alex Rives, who said that creating a virtual cell is "one of the most important challenges for the next era of science." According to Diamandis, Rives also said that today's datasets cover millions of cells and will need billions or trillions. Then Diamandis asked Wissner-Gross how long it would take before the US Food and Drug Administration (FDA) accepts drug testing on virtual cells.
Two steps: finding the drugs, then approving them
Wissner-Gross split the question in two. The first step is generating the candidates, and that is where his two-to-three-year forecast applies. He tied it to the math release, in which, in his telling, "substantially every problem of a certain level of difficulty" was solved "from one day to the next." He expects biology to have its own day N to day N plus one. He and Diamandis argue for this kind of industrial-scale problem solving in their online blueprint, Solve Everything, which also proposes infrastructure for testing candidates in the physical world, including shared robotic laboratories and clinical-device networks.
The second step, acting on those candidates, he called "a policy question" and "harder to predict." FDA-approved in silico clinical trials "hasn't quite happened yet," he said, though he sees the current agency inching that way, sometimes with single-trial approval routes and sometimes with Bayesian rather than frequentist statistics (two ways of judging trial evidence). The outcome, he said, depends on who is in the White House and who runs the FDA. His bet is that plausible candidates for thousands of diseases would create enormous political pressure. Either the FDA would move, or China would approve the drugs first and push the US into a competing system, producing "a total sea change in how we regulate clinical trials."
As an example of who might get there, Wissner-Gross pointed to Anthropic. He said the company had come out of stealth within the past month with a program to, "it appears, basically solve every single rare disease," running its own wet labs (labs for physical experiments) up to phase one clinical trials before handing candidates on. Anthropic's published announcements describe something narrower. In July it offered rare-disease research grants of up to $50,000 in Claude credits over six months for basic science and early-stage biotech projects, and it listed manufacturing capacity and safety testing as constraints. In September it described a Bay Area lab for fundamental molecular biology, where Claude agents searched genomes for candidate enzyme systems and human scientists then ran the experiments.
Wissner-Gross also said the virtual cell is only a start. He expects "a huge race" to scale up from virtual cells to tissues, organs and whole virtual organisms, and said a virtual organism could in principle tackle the top 5,000 diseases in one shot.
The case for speed: compress the data
Wissner-Gross's main argument for the timeline is an analogy with language. Ten or twenty years ago, he said, people called general cognition and natural language too complicated. Yet "all you had to do was just compress a large corpus": train a model on enough text with enough computing power. He also invoked the "bitter lesson," the argument that general methods plus computation beat hand-built expertise. He expects the same in biology. Take a big enough dataset, like the one Biohub and its partners are gathering, "just compress the darn thing," and the model will be forced to learn "what it takes to speak biology." "And if you speak biology," he said, "you can treat all disease."
Later in the episode, while talking about brain-computer interfaces, he extended the argument. He predicted that historians will see recent progress as "one big hairball." Compress dense brain recordings and a model can "speak brain"; compress proteomic and other multiomic data (measurements of a body's proteins, genes and other molecules) across a population and you speak biology; compress everything ever written, plus synthetic data, and you get a frontier math model. "All it takes is gathering huge data sets, compressing them with a lot of compute, and out pops the solution to everything." His case rests on this analogy and on the math release. In the episode he did not point to a biology model that had already done it.
Treatment as a route-finding problem
Diamandis raised the objection he said many people make: math and coding are confined problems, and biology is much more complicated. Wissner-Gross called that view "under-informed." Biology, he said, will be bulk solved "just as easily, if not more easily than math." It simply involves more data and higher dimensionality. Fermat's Last Theorem can be stated in a few tokens. A disease has to be described as a whole state of the body.
He sketched what that task might look like. Encode a patient's diseased state, combining physical traits with protein and epigenetic (gene-regulation) measurements, and treat a healthy state as the destination. Then plan a path of interventions between them. "That's just path planning. That's like AlphaGo," he said, predicting a "Move 37" for medicine once there is a digital twin of a whole human body. He compared it to using Google Maps: "take this pill, then take this pill," until the AI finds the route.
The references are to AlphaGo, Google DeepMind's Go program, which combined neural networks with search and reinforcement learning and beat Lee Sedol four games to one in March 2016. Its 37th move in game two looked strange at first but became part of a winning strategy. Go's legal moves and winning conditions are precisely defined. The objection Diamandis raised is essentially about whether the human body can be described that way.
Where the hosts pushed back, and where they piled on
Dave Blundin, founder and general partner of Link Ventures, agreed with the timeline but placed the difficulty somewhere unexpected. The hard step, he said, is from protein to cell, and "it will get cracked" because cells are not heavily regulated: wet labs can run millions of experiments to check whether a simulator is right. He argued that going from cell to whole organism is "not as hard as everyone would visualize," despite intuitions that an elephant is more than the sum of its cells. If the cell is cracked within three years, he said, the step to curing all disease is very short.
Diamandis added that labs already grow organoids: stem cells turned into miniature heart, liver, lung and kidney tissue, then connected so a drug can be tested across organs. Wissner-Gross described these as "the equivalent of synthetic data generation for post-training," with organoids producing the data instead of models.
Blundin and Wissner-Gross disagreed over how complex a cell is. Blundin noted that Google Earth amounts to about three petabytes and that "nobody quite knows the full complexity of the cell yet." Wissner-Gross replied that its complexity is known because the size of the human genome is known. Blundin countered that the genome gives only the base pairs, not the many proteins made from them. Wissner-Gross held his ground: "It can't be larger than the genome."
Salim Ismail, founder of Open ExO, said biology has always had "massive combinatorial complexity," so that the search space becomes far too large, and that AI can navigate it more elegantly. He asked whether pharmaceutical economics is therefore "cooked." Wissner-Gross said order matters: "the economy isn't commutative," and lucrative blockbuster drugs are still likely over the next few years. As evidence, he said GLP-1 weight-loss drugs bring in more revenue than OpenAI and Anthropic earn from all their frontier models.
Ismail then asked how long it would take to go from having the models to actually solving the diseases. "I think it'll happen overnight," Wissner-Gross said. Ismail said he has been telling audiences that cancer would be "cooked" within two or three years, and that this forecast goes further, to every disease. Wissner-Gross offered a way to check: "this is a falsifiable prediction. So we could check back in in a few years and see whether it happened or not."