By Titik Puspa 
Vijay Pande, a name once primarily resonating within the hallowed halls of academia, has undergone a significant transformation in his public perception, shifting decisively into the spotlight of the investment world. This evolution was dramatically catalyzed a dozen years ago when Andreessen Horowitz (a16z), a venture capital firm that had steadfastly avoided the healthcare and life sciences sectors for its initial five years, made a pivotal decision to embrace these fields. They entrusted the reins of this new endeavor to Pande, a distinguished Stanford chemistry professor renowned for his pioneering work on Folding@home, a groundbreaking distributed-computing project that harnessed the collective power of millions of home PCs to create a formidable supercomputer dedicated to disease research. Over the ensuing decade and more, Pande meticulously cultivated a16z’s investment in this area, growing it into a formidable practice managing nearly $4 billion in assets.
However, in a move that surprised many, Pande departed from this established success in June of last year to embark on a significantly more focused venture. His new firm, VZVC, co-founded with seasoned investor Zach Werner, is architected around a philosophy of making a select number of high-conviction, concentrated investments annually, a stark contrast to the broader, more diffuse investment strategies common in the industry. This new model eschews the traditional associate structure, relying instead on advanced artificial intelligence for its day-to-day operations.
To gain deeper insights into Pande’s deliberate pivot and the strategic underpinnings of VZVC, a comprehensive conversation revealed his rationale for championing concentrated bets in the current market landscape. He also delved into one of the most pressing conundrums facing AI-driven biotechnology: unlike text-based data, biological information cannot be easily scraped from the internet, leading to the creation of siloed, proprietary datasets by nearly every company. This raises critical questions about the future of AI-driven medical advancements and equitable access to their benefits. This conversation has been thoughtfully edited for conciseness and clarity, with a fuller version available for listening.
Pande articulates a fundamental shift in the paradigm of biology, moving it from a "science of discovery" to one that can be actively engineered. He explains that historically, drug development often involved a substantial element of serendipity. However, the advent of AI and machine learning has revolutionized this process by enabling computers to develop a sophisticated understanding of complex biological systems. This computational prowess allows for the identification of precise drug targets for specific diseases, the design of novel therapeutics, and even the optimization of clinical trials, which represent the most costly phase of drug development.
Addressing the notion that clinical trials are becoming more economical due to the increased use of synthetic data, Pande suggests this remains largely an aspiration. While acknowledging that AI has indeed reduced the cost and time required to reach the clinical trial stage, the sheer expense of conducting these trials—often running into hundreds of millions of dollars—continues to drive up drug prices. The success rate of drugs progressing through the three phases of clinical trials hovers at a mere 20%. When eight out of ten drugs fail, and each failure incurs costs in the hundreds of millions, the amortized cost per successful drug becomes exceedingly high. Pande emphasizes that these failures are rarely due to fundamental biological errors made by researchers. Instead, a significant contributing factor is the reliance on animal models, such as mice, for initial drug testing. These animal models, he points out, are often poor predictors of human responses. While AI models are not infallible, they promise a far more accurate prediction than animal models, and once they surpass this threshold, the potential for breakthroughs becomes truly exciting.
The subsequent and critical phase in drug development, Pande elaborates, is determining whether a drug is the "right drug for me." This leads to the concept of personalized medicine, more formally termed "precision medicine." Currently, when a patient presents with a non-trivial medical condition, physicians often resort to a process of educated guesswork, limited by the diagnostic tools available. They prescribe a drug, and if it proves ineffective, they move on to another, and then another. This trial-and-error approach is prevalent in oncology and many other therapeutic areas. Pande argues that a far more beneficial outcome would be achieved if the initial drug prescribed was the most appropriate one. Typically, patient blood test results are compared against population averages. However, a more insightful approach would be to determine if a specific result is anomalous for that individual. The emerging capabilities in medicine, particularly with AI, are enabling a deeper understanding of what treatments are best suited for each individual.
The journey to this transformative moment, Pande observes, is not the result of a single breakthrough but rather a convergence of multiple advancements. For a considerable period, precision medicine was primarily anchored in genomics. However, Pande draws an analogy: one’s genome is akin to the initial blueprint of a house, but over time, the house undergoes significant modifications and changes, making the original blueprint less representative of its current state. Consequently, advancements in measuring other biological molecules, such as proteins (proteomics), have become increasingly crucial for understanding disease progression and an individual’s current physiological status. This progress is intrinsically linked with the parallel advancements in automated robotic measurements, which, when integrated with AI, create a powerful synergy.
Over the past decade, there has been a consistent and significant pace of innovation in both AI for biology and AI for chemistry. The biological aspect focuses on identifying the most effective therapeutic strategies for treating diseases, while the chemical aspect aims to design drugs that can precisely target specific proteins implicated in those diseases. These two intertwined fields have witnessed substantial progress over the last ten years.
Pande highlights a unique challenge in biology as a domain where AI cannot simply "scrape data off the internet" in the same way it can with text. This characteristic has profound implications for the field’s development. It signifies an environment where readily available, universally applicable training data for AI models is scarce. Unlike text-based AI, where vast repositories of online information can be leveraged to train models, biological data is largely proprietary and often siloed. This presents a fascinating challenge from a pure AI perspective, as it necessitates novel approaches to data acquisition, curation, and model development.
This data scarcity, Pande acknowledges, echoes a persistent issue in medicine: the tendency for medical professionals to operate within distinct, often competitive, silos. He elaborates on this point, suggesting that AI holds the potential to bridge these divides. For instance, if a patient’s condition involves both oncological and endocrinological complexities, these specialists may not always collaborate seamlessly. AI, however, can theoretically function as a universal specialist, capable of synthesizing information and identifying patterns that might elude any single human expert. This could effectively create an advisory council of the world’s leading medical minds, all converging on a single patient’s case.
The realization of this vision, however, hinges on the extent of data sharing within the medical community. While founders and investors understandably seek to protect their proprietary findings, Pande points to a significant trend towards the development of comprehensive "atlases" of biological information. These atlases, from a technological standpoint, are often built upon foundation models. As these models become more prevalent, Pande anticipates a phenomenon similar to the rise of open-source Large Language Models (LLMs), which have demonstrated remarkable capabilities and often outperform their proprietary counterparts. He believes that open-source foundation models in biology will exert a similarly broad and transformative impact.
Pande’s current investment focus includes companies like Genesis Therapeutics, which originated from his lab at Stanford, and Insitro, a pioneering drug discovery company founded by his former Stanford colleague, Daphne Koller. He is also actively incubating a new venture with a founder he has known for two decades. When asked about his criteria for selecting founders and the areas he is targeting, Pande identifies two primary domains: AI for healthcare delivery and AI for clinical trials, areas he also extensively explored during his tenure at a16z.
Central to his founder selection process is the paramount importance of mutual trust. He seeks founders who exhibit high integrity and consistently deliver on their commitments. Pande envisions his relationships with founders as long-term partnerships, ideally spanning five to ten years and potentially extending to their subsequent ventures. He values founders who adopt a long-term perspective, prioritizing collaborative success over mere competitive advantage.
Reflecting on his investment career, Pande identifies both successes and learning experiences. He recalls the initial resistance and skepticism he encountered over a decade ago when he first began advocating for the integration of AI and machine learning into medicine and biology. Many dismissed these concepts as improbable or impractical. The subsequent widespread acceptance and the visible arc of progress in this field have been deeply gratifying.
He admits that it took him time to fully appreciate the critical importance of go-to-market strategies, even when faced with the allure of cutting-edge technologies. Pande now advises his founders, particularly those with strong scientific or product backgrounds, to dedicate their considerable brilliance and creativity to developing robust go-to-market plans, recognizing that this aspect is often as challenging, if not more so, than the technological innovation itself.
Pande offers a detailed explanation of how his new firm, VZVC, is designed differently from his previous role at a16z. The firm, named after Pande (Vijay) and his co-founder Zach Werner (Z), is intentionally structured to be exceptionally lean. On the investment side, the core team consists solely of Pande and Werner. While they initially considered hiring associates, the advanced AI tools they have developed have rendered this unnecessary, allowing them to operate with a highly efficient structure.
The level of concentration in their investment strategy is notably high. Instead of making 30 bets per year, VZVC aims for approximately five highly concentrated investments. Pande likens the typical fund’s approach to adding a new Facebook friend—a relatively quick decision. In contrast, for Zach and himself, making an investment is a far more significant commitment, akin to the decision to have another child.
With this concentrated approach, VZVC often finds itself not directly competing for hot funding rounds. Instead, companies tend to make room for them as investors, recognizing the unique value Pande and Werner bring through their hands-on involvement and expertise. Pande cites Antonio Gracias of Valor Equity Partners as an inspiration, whose long-standing success, exemplified by his involvement with SpaceX, predates his current prominence. He also points to Thrive Capital’s strategy of maintaining a more concentrated portfolio as another significant influence. While a16z remains deeply embedded in his professional DNA, he views Gracias and Thrive as important new paradigms influencing his current thinking.
When asked about what is currently overhyped in AI and biotech, Pande cautions against the narrative that AI will be a panacea for all medical challenges. He emphasizes that while AI can indeed uncover insights beyond human capacity, the primary limitation lies not in AI’s potential but in the availability and quality of data. Large Language Models, for instance, achieve their remarkable performance due to the vast amounts of data they are trained on. In biological applications, where such extensive and comprehensive datasets are not readily available, AI cannot magically overcome these data limitations. The true challenge, therefore, lies in bridging the data gap, rather than solely in the advancement of AI algorithms themselves.
