Guides · For company owners
Selling company data to AI labs: what we've learned
Owners get locked in early, buyers chase samples across five apps, and data gets underpriced. Here's what we've learned and what to ask any broker.
Erik Newsham · · First published on X
Short version
The market for real company data is young, messy and moving fast. Three problems come up again and again: owners tied to a single broker before they've seen a real offer, buyers chasing samples and prices across five apps, and owners not being paid what their data is worth. If you're selling, get several offers, insist on a real price before you commit, and ask exactly what you'll receive.
The strangest market I've worked in
For this market to exist, people who have spent years building a business have to hand their records to some of the largest technology companies in the world. If I were watching from the outside, I'd have shaken my head and stayed well away.
Instead my life has been consumed by it. Our team works very long days talking to business owners, AI labs, data handling specialists, partners, private equity firms and, yes, governments. Hundreds of companies come to us every week wanting to sell their data, and buyers are asking us for more than we can find.
Here are the three biggest problems we've seen, and what we're doing about each.
Problem 1: Owners getting locked in and missing their window
This may cost us business, but it's true: if you want to sell company data, the best thing you can do is talk to as many brokers as you can and collect as many prices as you can.
The early rush is fading. The supply of company data is growing quickly, and the bar a dataset has to clear keeps rising. The worst outcome for an owner is to depend on one broker's word, sign something that ties them up before a real offer arrives, and then miss the chance to license their data at all.
So get an evaluation from us, and get one from others too. Then compare the terms and the prices. A broker who's confident in their buyers won't mind.
When you compare, ask each one:
- Will I see a price from a real buyer before I commit to anything?
- Exactly how much will I receive, and when?
- Is the deal exclusive or non-exclusive, and for how long?
- Who will the data go to, and what can they use it for?
- Can I leave out systems, folders or records I don't want shared?
Our guide on how to sell your company's data to AI companies covers the full process and what to check before you sell.
Problem 2: Buying data is slow and inefficient
Buyers are tired of working with brokers (people like me) who send samples, prices and terms across Slack, Signal and late-night video calls. Every dataset starts with the same questions: what's actually in it, what does it cost, and can I see some of it?
So we built Grepped Buyside. We picked a few hundred of our company data collections and gave each one a detailed description, an indicative price and a sample a buyer can open straight away. It's open to our invited lab partners now, and we're letting in RL environment teams and newer labs from the waitlist.
If you build RL environments, real company data for RL environments explains what's available and what to check before you buy.
Problem 3: Owners not getting their data's real value
There are two failure modes here.
The first is overpromising. A lot of cold outreach promises huge demand for a ten-person company's data. Honestly, we rarely take on a ten-person company unless it's in an unusual sector with records across several linked systems. We'd rather say no than promise something we can't deliver.
The second is underpaying. We've heard stories of company data bought for $50,000 and resold for millions. That isn't fair on the owner who built it.
Our answer is to be clear with owners about what they'll receive before they commit, and to set prices with the owner rather than for them. If you want to understand what drives the number, read what your company's data is worth to AI labs.
Where this goes next
This is still the strangest market I've seen. But I don't think it has to stay that way. Owners should be able to compare offers and get paid what their data is really worth. Buyers should be able to find what they need without chasing anyone across five apps at midnight. We're building that. We're not there yet, but it gets a little better every week.
- If you own a company and want to know what your data is worth, talk to us. Then talk to everyone else too.
- If you're a broker struggling to sell what you have, or you can reach dozens of companies through private equity, venture or an industry network, apply to our partner program.
- If you build RL environments or train models, join the Buyside waitlist.
And to the few thousand owners, partners and labs who've already taken a chance on us: thank you.
Find out what your data is worth
Tell us what you hold and we'll come back with an estimate and the buyers who want it.