Vijay Pande spent more than a decade turning a single a16z experiment into a life sciences practice managing close to $4 billion. Then last June, he walked away to build something deliberately tiny. According to TechCrunch AI, Pande’s new firm VZVC, co-founded with longtime investor Zach Werner, runs on a handful of concentrated bets a year, employs no associates, and leans on AI for daily operations.
What stands out here is the direction of travel. Most investors who leave a megafund raise a bigger one. Pande went the other way. “We’re not doing 30 bets a year,” he told TechCrunch AI, and that line is the whole thesis. In a crowded, capital-soaked AI market, he’s betting that fewer, deeper positions beat spraying money across dozens of startups.
Why go small now?
The timing matters. When capital is cheap and everyone is chasing the same AI deals, spreading thin means paying up for mediocre entry points. A concentrated fund with no associates and AI handling the grunt work can move faster and stay picky. It’s a structural answer to a market problem: too much money, not enough conviction.
There’s also a signal for founders and operators. A firm that runs its own back office on AI is putting its money where its thesis is. If the tools are good enough to replace an analyst bench, that’s a live demonstration, not a pitch.
What’s actually changing in AI biotech?
Pande’s bigger argument is that biology is shifting from a “science of discovery” to something you can engineer. For most of drug development, finding a working drug had a lucky, fortuitous quality to it. AI changes the odds by helping pick targets, design the drugs, and now even shape clinical trials, which remain the most expensive part of the process.
The numbers explain the urgency. Pande told TechCrunch AI that a drug’s probability of going from the first trial to the end of the third is just 20%. When 8 of 10 fail and each trial can cost hundreds of millions, the amortized cost is brutal, and that’s why drugs are so expensive.
His explanation for the failure rate is the interesting part. Drugs usually fail not because a biologist erred, but because they were designed on animal models like mice, and mice just aren’t very predictive of humans. “The AI model is not going to be perfect, but it’s going to be way better than any animal model would be,” he said. Cross that bar, and the economics of drug development start to bend.
The data problem nobody can scrape around
Here’s the conundrum Pande keeps circling. Unlike text, biological data can’t be scraped off the internet. Every company ends up building its own walled-off dataset, and that data can’t be cheaply distilled from one model to another. From a pure AI standpoint, it’s a fascinating setup. From a public-benefit standpoint, it’s a warning.
If the promise of AI in medicine is precision, treating what’s actually weird for you rather than comparing your bloodwork to a population average, then whoever controls the datasets controls access. Pande also flagged the silo problem inside medicine itself, where an oncologist and an endocrinologist treating the same patient “really don’t sync together very well.” AI could, in principle, be a specialist in everything at once, like having a team of the best doctors in the room. But only if there’s enough data to feed it.
What to take from this
- Watch fund structure, not just fund size. Lean, AI-run firms making concentrated bets are a bet on operator quality over deal volume.
- In AI biotech, proprietary data is the moat. Ask who owns the dataset before you value the model.
- The near-term unlock isn’t cheaper trials tomorrow. It’s AI models that beat animal models on predicting human response.
Pande built a $4 billion practice by reading a category early. Now he’s reading his own industry the same way, and the read is that smaller and sharper wins the next cycle. You can find the fuller conversation at the original source.