So, You Want to Start a Neolab
What we've learned after seeing hundreds of these decks
We see a lot of neolab decks.
By last count, almost 500 so far.
And the founders pitching them are, almost without exception, brilliant. Technical pedigrees. Previous experience. Many have spent years at frontier labs.
We love that energy. We want to fund it.
Here’s what we actually look for.
1. A Novel Approach
Look, I’m sorry. You’re not going to be able to build an LLM better than Claude or Chat. Those are incredible teams, backed by billions of dollars.
Instead, you had better have a novel approach.
This sometimes get called Alternative Architectures, but really, you need to have something that distinguishes you in the race, and that could give you a legitimate (and ideally durable) leg up over others.
This could be things like novel interpretability techniques, formal verification, world models. Whatever, but it needs to be different and it needs to be real.
Ideally, you also have access to something proprietary. That could be a research edge, or data that is not public or something else, but in a landscape with as much capital and competition as AI has, it’s never been more important to have a clear story about an unfair advantage.
2. Serious Credentials
This category is unforgiving. You need people on the team who have done the work — not adjacent to the work, not inspired by the work, but actually done it. Published the papers. Trained the models. Shipped the systems.
There’s a funny lag right now around the fact that there are folks who have been working on AI for 30+ years, but who just recently have found their skills incredibly highly valued by the market.
The bar is high because the technical risk is high. Investors know that and they’re looking for signal that the team can actually execute on the technical thesis, not just articulate it.
3. A CEO Who Can Fundraise
This is in some ways super obvious and somehow also incredibly underrated.
Sam Altman does not know how to build AI - but he’s an excellent fundraiser.
Dario maybe knows a little more than Sam, but he’s no research genius, he’s an excellent fundraiser.
Demis Hasabis is probably the most sophisticated technical CEO, and that’s why it was probably good he sold to Google so early - he stopped needing to fundraise.
The very frank reality is that AI is expensive in ways Saas was not.
You will need lots of money in order to be able to succeed, and you will need to be able to keep investors excited in you for potentially long, drawn out stretches as you work to make progress.
You need a CEO who can walk into a room with a sovereign wealth fund, a strategic, or a top-tier VC — and close. Not just impress. Close.
Technical credibility in the CEO seat is good. But ultimately, your CEO must fundraise at scale, or you will run out of runway.
4. A Clear Market and ICP
Here’s what separates the pitches that get us excited from the ones that don’t:
You must have a clear idea of who you are building for.
This is startup 101, but far too many wannabe labs come to me with a technical idea and absolutely no thought whatsoever to a customer profile.
I will say this clearly, you must be able to answer: who will pay for this, and why will they pay for you specifically?
“We’re going to build a better model” is a grant application. “We’re going to build a better model to solve X for Y customer” is the beginning of a business.
Neolabs who start with a market application and are then led to technical differentiation are exciting. You want a novel approach because there’s a specific problem worth solving, not the other way around.
The GTM thesis doesn’t have to be fully baked. It can be a hypothesis. But it has to be there, and it has to be specific enough to test.
You can almost think of it like a brand positioning statement:
“We have a novel approach to [specific capability], we believe it unlocks [specific application], and we’ve already had [early signal] from [type of customer]”
This is infinitely more fundable than the team that says “we think foundation models are underrated and we can build a better one.”
The Honest Summary
Neolabs are cool. The world probably needs more of them, not fewer.
The frontier labs have their incentives and their constraints, and there’s real room for focused, well-credentialed teams to do things the big labs can’t prioritize.
But it’s a hard category to invest in, and founders who come in clear-eyed about what they’re actually building — and for whom — will always get further than the ones who are just betting on talent and a big idea.
If you think you have all three legs of the stool, and you have a customer in mind, we want to hear from you.
Originally published on Griff’s Substack on 29 May 2026.