Veteran Biotech Investor Abandons Mega-Fund Model for Concentrated Bets

Vijay Pande left a16z's $4 billion life sciences practice to launch a lean venture firm betting on AI's potential to transform drug development.

By Duabbly Staff

Vijay Pande spent more than a decade building Andreessen Horowitz's life sciences division into a nearly $4 billion operation, transforming himself from a Stanford chemistry professor into one of the venture capital world's most influential investors in medicine and biology. Yet in June 2025, he walked away from that sprawling portfolio to launch something radically different: a new firm called VZVC, co-founded with longtime investor Zach Werner, built around a handful of concentrated annual investments rather than dozens of smaller bets.

The shift reflects a broader reckoning among some venture investors about the pace and scale of capital deployment in biotech. Where the traditional venture playbook calls for spreading risk across numerous companies, Pande's new model prioritizes depth over breadth. The firm eschews the typical associate structure and leans heavily on artificial intelligence for operational tasks that would traditionally require larger teams. It is a stark departure from how Pande operated at a16z, where he managed one of the largest dedicated biotech practices in the industry.

When Andreessen and Horowitz first launched their firm roughly fifteen years ago, they deliberately sidestepped healthcare and life sciences entirely. The calculus changed as the venture landscape evolved and as Pande's track record grew. His background was unconventional for venture capital. Before a16z recruited him, Pande was best known in academic circles for creating Folding@home, a distributed computing project that harnessed millions of personal computers worldwide to model protein folding and conduct disease research. That project demonstrated his capacity to think at scale about biology and computation, a combination that would prove valuable as AI began reshaping drug discovery.

The transition from academic researcher to venture investor, then from mega-fund investor to focused founder, reflects Pande's evolving thinking about how medicine actually gets made. In recent interviews, he has articulated a vision of biology shifting from a discipline rooted in discovery and serendipity to one that can be engineered with precision. This transformation, he argues, hinges on artificial intelligence and machine learning's capacity to process biological complexity in ways human intuition cannot match.

Traditionally, drug development has relied on educated guesses and fortunate accidents. Researchers identify a disease, run experiments on animal models, and hope the results translate to humans. The track record on that final translation is dismal. Only roughly 20 percent of drugs that enter human clinical trials ultimately receive approval, meaning eight out of ten fail. Those failures rarely indicate scientific incompetence; rather, they expose the fundamental limitation of animal models as predictors of human response. When a single clinical trial can cost hundreds of millions of dollars, that failure rate drives up the amortized cost of successful drugs dramatically, creating one reason why pharmaceuticals remain prohibitively expensive for many Americans.

Pande sees artificial intelligence as offering a way to collapse that probability gap. AI models, trained on biological data, can predict human response with greater accuracy than mice or other animal surrogates. Once AI predictions exceed the predictive power of animal models, which Pande believes is already beginning to occur, the economics of drug development shift fundamentally. The cost and timeline to reach human trials should compress, and failure rates should decline. That cascade of improvements could reshape the entire industry.

Beyond identifying which drugs might work, AI is enabling a second transformation in medicine that Pande calls precision medicine, though the field typically uses the term precision or personalized medicine. Currently, when a patient visits a doctor with a complex condition, diagnosis often involves guesswork followed by sequential drug trials. The patient receives a medication; if it fails, they receive another, and so on. This trial-and-error approach persists in cancer treatment, autoimmune diseases, and countless other conditions. The alternative would be identifying the right drug for that specific patient on the first attempt, something that requires understanding not just what the population average looks like but what normal means for that individual.

Much of precision medicine historically relied on genomics, reading the patient's genetic code as a kind of biological blueprint. Yet Pande emphasizes that a genome resembles a house's original blueprint, not the house as it exists today. The body changes constantly. Measuring proteins through proteomics and other emerging techniques provides more relevant information about a patient's current state than their genetic code alone. Combined with advances in robotic measurement and automation, which naturally integrate with AI systems, these tools enable doctors to tailor treatment to individual biology rather than population statistics.

The convergence of these technologies did not happen suddenly. Over the past decade, parallel advances in AI for biology and AI for chemistry have accumulated steadily. The biological dimension addresses which diseases to target and how to intervene. The chemistry dimension solves the molecular problem of designing a drug that hits a specific protein. Both threads have matured considerably, even if the public has not widely noticed.

One challenge unique to AI-driven biology distinguishes it sharply from other AI applications. Text, images, and video pervade the internet, allowing AI researchers to train models on massive datasets scraped from the public domain. Biological data cannot be harvested from the internet. Nearly every biotech company building AI systems for drug discovery must construct its own proprietary dataset, a walled-off resource unavailable to competitors or the broader research community. This creates a situation that echoes an old problem in medicine itself: researchers and clinicians operating in territorial silos, hoarding information rather than pooling knowledge.

The data question shapes how Pande and Werner think about the companies they back. Unlike biotech firms of the past, which could compete primarily on scientific insight or chemical intuition, future winners will likely be those that accumulate the richest biological datasets and build the most powerful AI systems to interpret them. That concentration of data advantage may entrench winners in ways that benefit patients receiving better treatments but also create formidable barriers to entry for challengers and limit the democratization of medical AI that some advocates envision.

Pande's decision to scale down his fund size and concentrate his bets on fewer companies reflects confidence in this emerging landscape. Rather than hedging risk across dozens of investments, VZVC will bet heavily on companies positioned to accumulate data advantages and deploy AI effectively. Whether this model succeeds depends partly on whether the technological promises hold up and partly on whether regulatory and clinical systems can keep pace with innovation. For now, Pande is wagering that they will, and that smaller, more focused investment can move faster than sprawling mega-funds in backing the future of medicine.