Data beyond brokerage

Data beyond brokerage

Everyone else is getting rich selling data.

Yoni Rechtman

Oct 09, 2026

15

4

1

Share

Everyone else is getting rich selling data. It’s never been more over.

Fortunes are being made overnight feeding the voracious, seemingly endless demand and budgets for data from the model labs, well funded AI app companies, and robotics labs. The data business is probably the single most lucrative opportunity in the world right now. There are just massive dollars flowing around up for grabs.

Serving this may not be a good business with terminal value. Right now it seems like buyers have basically unlimited budgets for the data that they want. And it stands to reason that those budgets will grow for some time.

That

might

be a temporary condition, and the eventual cost pressures from being public make it more likely to be that way. Or if it does last, it seems likely that “what they want” will change/turn over often enough such that

even if you have “the right tokens” today you’re unlikely to have them tomorrow

. So macro, data demand may dry up and/or get reigned in over time, and micro it is likely to change substantially very quickly to your detriment, probably multiple times.

So if you’re going to chase this money, you probably need to plan on doing that either as a short lived, high FCF, low TV trade (an S-Corp) or as a way to fund something else.

Accounting for Data

Businesses can fund themselves through debt, equity, or cash flows. Now data sales are this new source of funding to mix in (FCF for cash-burning companies). Think of data sales as

a new source of capital rather than a potential revenue stream

.

Data sales technically show up as revenue, which is great, but the opportunity for startups is not to build “we sell data” as a company. Rather, “we sell data” is this great point in time opportunity to go and do something else that may have real terminal value. If data sales are an exhaust from some primary thing, it makes it much more economically viable to climb that hill. If you get the model-labs-monopoly-money, use it to fund something rather than as the goal unto itself.

If there’s not really natural or organic ways for you to get or generate data (broadly that might be seed data, synthetic data, RL environments, etc.) then it costs you focus and resource allocation, and ultimately

bends you to the roadmap of a customer who might go away tomorrow

.

For the moment, founders should obviously talk/think about data sales as revenue because it’s what the market wants to hear. But there is clearly some work to do on accounting standards here as to how data should be treated on a balance sheet. Right now, selling data basically creates money from nothing. You sell an asset that was never accounted for -

manna from heaven

.

Companies are very likely trading at valuations that don’t account for the values of their data.

We have a number of companies working on this in different areas, using or contemplating data sales to fund go to market for other products/services/opportunities across consumer apps, agents, and asset heavy businesses. See also, the

Context Acquisition Company

(more to come soon).

The alternative is to be in a high cash flow trade of TBD duration, in which case you should keep an eye on the exits. Look to get cash out of those businesses either through dividends or a strategic acquisition. I don’t think there’s a ton of TV in data brokerage, BUT they are clearly so cash flow generative and so strategic that they get funded well and acquired easily. Someone will buy Mercor for a lot of money, both for its AI talent, its data, to impair competitors, etc.

So:

are you doing X to sell data, or are you selling data to fund X?

In the short term they may look the same.

Shelby wrote really well about thi

s.

Data is non-rival. Once you sell it the lab has a copy forever, so when does selling your exhaust mean selling your moat? It’s highly contextual.

For AI apps businesses, it probably does not make sense. Harvey should never sell data about legal workflows to the models. That data is its prime differentiator and eventual source of margin. But if you own or are monetizing some other asset which either generates or could generate valuable training data, that seems less likely to be a risk. There’s also a difference between historical data and production data. If you operate a business/asset, you could sell the data once while continuing to accumulate / create new data through usage.

New business models with data sales

I’m really interested in anyone who has thoughtful ideas on building businesses that can either make great use of data sales or generate data opportunities for themselves as “exhaust” from their regular operations. Some places that could happen:

Basically any company with broad consumer reach should at least be considering data sales as a way to fund expansion (paying people/generating tasks and data based on the distribution you have). This is especially true/tantalizing in games and entertainment, where you can generate world model data.

Same goes for really any company with physical operations; strap a camera on and start getting some ego data!

Anyone who wants to use data to offer price cuts or even pay customers to use services.

This is likely how data emerges as a new business model akin to advertising

.

Probably the most interesting set of companies are those who can actually get PAID to acquire data. AI transformation companies can likely be great partners to their customers (who already pay them) that want to get into data sales, especially as they accumulate access to data from non-core parts of their business (does an industrial co really regard its accounting practices as trade secret?). We’ve met a lot of companies working on this within screen recording and computer use data specifically.

The point in any case is that data sales can either create opportunities to capture value from a business you’re already building or accelerate you in building it. I’m relatively less bullish on straight up brokerage businesses that don’t have a theory on some compounding advantage to create durable/terminal value.

To the extent data sales themselves are a lasting business with terminal value, it’ll be through new capital structures and business models rather than rerunning web 2.0 and SaaS models.

Join us for dinner

Next month Leeor Mushin and I are hosting a dinner in NY for data brokers and buyers. Hit me up to attend or recommend someone who I should invite. This will be a small group of ≈10 people.

15

4

1

Share