The man who wrote $4 billion checks at Andreessen Horowitz is now writing smaller ones — and that might tell you more about where AI is actually headed than any mega-round announcement.
The Summary
- Vijay Pande left a16z's $4B biotech practice to launch VZVC, an AI-native fund making fewer, more concentrated bets
- Biology is transitioning from "discovery science" to "engineering discipline" — meaning predictable, repeatable design rather than lucky finds
- Open datasets, not proprietary data moats, will unlock AI's real value in medicine
- Clinical trials remain the constraint: even perfect AI drug design hits the same decade-long, billion-dollar validation wall
The Signal
Pande's move isn't about checking out. It's about focus. At a16z, he built one of the largest biotech funds in venture. Now he's running something deliberately smaller, making what he calls "not 30 bets a year." The thesis: AI in biology is real, but most investors are funding the wrong layer of the stack.
The shift he's betting on is fundamental. Biology has always been empirical — you discover things through experiment, observation, trial and error. Engineering is different. You design from first principles. You predict outcomes before you build. Pande argues AI is finally giving biology that predictability.
"Biology is shifting from a discovery science to an engineering one."
What does that mean in practice? Instead of screening thousands of molecules hoping one works, you design the molecule you need and it works the first time. Instead of running experiments to see what a protein does, you simulate it and know before you synthesize it. The cycle time collapses. The capital efficiency changes. This isn't incremental. It's categorical.
But here's where Pande parts ways with the current AI hype cycle: he thinks the real unlock isn't proprietary models or walled-off datasets. It's open, shared data. The companies hoarding patient records or molecular libraries are building moats in the wrong place. The constraint in medicine isn't data scarcity — it's data accessibility and interoperability.
AI models get better when they train on diverse, high-quality datasets. In medicine, those datasets are fragmented across hospitals, research institutions, and private companies. The startups that win won't be the ones with the biggest private trove. They'll be the ones that figure out how to aggregate, standardize, and learn from distributed sources without centralizing control. That's a Web3-adjacent problem, even if Pande isn't framing it that way.
Here's the cold water: even if AI nails drug design, clinical trials are still a decade-long slog. You still need to prove safety in humans. You still need Phase I, II, and III trials. You still need regulatory approval. The FDA doesn't care how good your model is. They care about outcomes in actual people. That's a billion-dollar process, minimum.
So the bottleneck isn't discovery anymore. It's validation. AI compresses the front end — the years spent finding a drug candidate. But it doesn't compress the back end. Not yet. The companies Pande is betting on have to navigate both.
The Implication
If biology is becoming engineering, then the next wave of biotech looks less like Genentech and more like Intel. Repeatable processes. Design tools. Platforms, not one-off products. Watch for startups building infrastructure — simulation engines, protein design tools, clinical trial optimization software — not just novel drug candidates.
And if open datasets matter more than proprietary ones, then the AI biotech winners might not be the ones with the most compute or the deepest pockets. They'll be the ones who figure out interoperability, data standards, and collaborative architectures. That's where the real edge is.