Why Vijay Pande Left a $4 Billion VC Practice to Bet Small on AI and Health

Vijay Pande, the Stanford chemistry professor who built Folding@home into one of the world's largest distributed-computing projects, left Andreessen Horowitz in June 2025 after growing the firm's healthcare and life sciences practice into a portfolio managing close to $4 billion. He has since co-founded VZVC with longtime investor Zach Werner and is currently raising the new firm's first fund, which will focus on AI and consumer health. (TechCrunch, STAT News)

VZVC's structure is deliberately lean. The firm makes a handful of concentrated bets per year rather than the dozens typical of large venture funds, employs no associates, and relies heavily on AI for its day-to-day operations. Pande described the approach plainly: "We're not doing 30 bets a year." (TechCrunch)

The contrast with his previous platform is stark. Marc Andreessen and Ben Horowitz avoided healthcare and life sciences for a16z's first five years before committing to the category and handing it to Pande, who built the practice from scratch. a16z's bio and health fund last closed in 2022. After Pande's departure, reported on June 10, 2025, a16z began seeking a much smaller $750 million fifth fund for its bio and health strategy. (TechCrunch)

Pande's thesis for VZVC rests on a structural argument about how biology itself is changing. He said biology is moving from a "science of discovery" — where researchers find what nature happens to produce — to something that can be engineered, with AI helping identify drug targets, design molecules, and assist in clinical trials. The stakes are concrete. Running a clinical trial can still cost hundreds of millions of dollars. The probability of a drug successfully progressing from its first trial to the end of the third trial is about 20%, meaning 8 out of 10 drugs fail. Drugs typically fail, Pande said, because they were designed on experiments using animal models like mice, which are not very predictive of human outcomes. (TechCrunch)

He also pointed to what he called "precision medicine" as an area where AI could shift the equation. Patients today are often given one drug after another when the first does not work, a trial-and-error process that better predictive models could compress. (TechCrunch)

The structural conditions Pande describes are not speculative. The 20% Phase I-to-Phase III success rate he cites has been a known constraint in pharmaceutical development for years. The cost figures for clinical trials are consistent with widely reported industry estimates. What is new is the framing: Pande is arguing that AI can compress the drug-development pipeline not at a single point but across target identification, molecule design, and trial execution simultaneously.

Whether AI can meaningfully move that 20% figure is the open question. Machine learning models for predicting molecular properties have improved, but drug candidates still fail in late-stage trials for reasons that are often biological rather than computational: toxicity that animal models did not predict, effectiveness that early-phase trial populations did not replicate at scale, or the sheer variability among patients that no model fully captures. Pande's own description of the problem centers on the inadequacy of mouse models, which suggests he is betting on AI-driven modeling that is more relevant to human biology, not merely faster screening.

VZVC's operational design is itself a data point on how AI may change venture capital. A firm with no associates and heavy AI reliance for day-to-day operations implies that deal sourcing, due diligence, portfolio monitoring, or some combination of these functions are being automated or augmented to a degree that would have required a larger team a few years ago. The concentrated-bet model means each position carries more weight, which raises the cost of a wrong call but also concentrates upside. For a firm focused on AI and consumer health, the thesis and the operating model are aligned: both assume that AI compresses cycles, whether in drug development or in the investment process itself.

Pande's track record gives the thesis weight. Folding@home demonstrated that distributed computation could attack biologically meaningful problems at scale, using idle processing power from home PCs to simulate protein folding — the process by which a protein chain assumes its functional shape. That project anticipated, by more than a decade, the current intersection of computation and biology that firms like VZVC are now trying to monetize. The question is whether the gap between simulation and clinical success has narrowed enough to justify concentrated bets on companies operating in that space.

VZVC has not yet closed its first fund. The firm's focus on AI and consumer health positions it at the intersection of two categories that have attracted significant capital but produced uneven results. Consumer health ventures have historically struggled with reimbursement, regulatory pathways, and distribution. AI-driven drug discovery has produced high valuations but limited late-stage clinical validation to date. Pande is betting that the combination, pursued through a concentrated portfolio and an AI-native operating model, can find value that broader, more traditional approaches have missed. (STAT News, Endpoints News)