somewhere in the last couple of years the better buyers in vertical software quietly changed what they were buying.
they still say “distribution” when they talk about go-to-market, but if you look at what they actually write checks for, they’ve started treating it as a thing you own.
something you buy, hold, and price. in a few cases they’ve paid more for the distribution than for the business itself.
i didn’t start out with a theory here. i started with a list of deals that didn’t make sense under the usual logic, and the theory is what was left once i tried to explain them.
the tape
march 2026. Beacon Software pays eight figures for RealEstateAPI, a bootstrapped property-data company. about a dozen people, 300 paying customers, no venture money, no institutional backing of any kind.
that’s an odd purchase for Beacon. they’re a rollup. their whole model is buying operating businesses with recurring service revenue, and RealEstateAPI isn’t one. it’s a data layer: structured property data on roughly 150 million american homes, sold programmatically to investors, insurers and lenders.
the customer contracts alone don’t justify the price, and Beacon knows how to value customer contracts better than almost anyone.
so my read is they were paying for something else. if you consolidate enough property managers, you eventually end up holding a view of residential real estate that institutional buyers would pay for.
RealEstateAPI is that view.
Beacon bought the reach, and what the reach can see - and got a small software business thrown in.
may 2026. Constellation Software acquires Keypoint Intelligence. Keypoint is the benchmark house for digital imaging, the reference point the whole vertical prices and compares itself against.
Constellation has done hundreds of acquisitions and has a reputation, deserved as far as i can tell, for never overpaying and never straying from the thesis: niche vertical software, recurring revenue, hold forever.
Keypoint isn’t really software. it’s a benchmark business. i find it hard to read that deal as anything other than Constellation deciding that the benchmark position, the place the market routes its decisions through, now counts as the kind of thing they buy.
april 2026. OpenAI makes what i believe is its first media acquisition: TBPN, a daily tech talk show with fifty-eight thousand youtube subscribers, for a reported price in the hundreds of millions.
per subscriber the number is silly. per room reached (the specific people in that audience, and who they talk to) it’s arguably cheap. the company whose models everyone else rents decided that the channel was scarcer than the model. i don’t think that was a marketing decision.
none of this is entirely new. HubSpot has been buying audiences for years (The Hustle, Starter Story). Clay paid something like $200 a subscriber for Clay Nation. the shape repeats: a buyer with a product that scales at near-zero marginal cost, lacking the room to sell it into, paying roughly what it would have cost to build that room themselves. replacement cost, basically.
and if you go back further and up a weight class, you get the deals that the current ones rhyme with.
Verisk, which began as an insurance data consortium and is now a listed company, is maybe the end state of the whole thing: the distribution outlived every product that was ever built on top of it.
you can call these marketing spend if you want. the buyers don’t. they’re carried as assets, held as assets, and so far they’ve appreciated like assets.
what is actually an asset class?
i want to be careful here, because “asset class” is a phrase people reach for when they want something to sound institutional.
the conventional test is something like: a distinct risk profile, a distinct economic driver, and returns that don’t just track equities or bonds. i think distribution, priced the way the deals above price it, mostly passes. mostly.
the risk is mainly time. a panel that starts recording in 2026 can’t be caught by one that starts in 2028, because history only runs one direction. the threat to a distribution position isn’t a competitor with a better product. it’s being late. every month you don’t hold the position is calibration you don’t get back. real estate can be built, equity can be issued, but once someone else holds the position in a vertical, there isn’t a second one to buy. i can’t think of another asset where lateness is the dominant risk in quite this way. (i’m open to being told one.)
the driver is density rather than revenue, and it’s non-linear. fifty nodes in a vertical is worth close to nothing. fifteen hundred nodes in the same vertical can be worth eight figures. below some threshold the panel effectively doesn’t exist as a panel; above it, it starts to appreciate on its own. that’s different from a software business, where value tracks cash flow more or less continuously.
the returns are the part that trips people up. they don’t come from the underlying operation. Beacon’s RealEstateAPI returns come from institutional buyers of property intelligence, not from property managers. Keypoint earns because the imaging industry prices itself against Keypoint’s benchmarks, not because Keypoint does anything in imaging. the buyer for the asset is a different class of buyer than the customer of the business the asset was built on. equities earn from a company’s own customers, bonds from its own cash flow. this earns from a market’s willingness to pay to understand itself. and unlike most things in software it gets more valuable with time, not less.
where the framework is weakest, honestly, is liquidity and comparability. there’s no market for these positions yet, only one-off transactions, and the prices are noisy. i think that changes, but it hasn’t yet.
why now
AI pulled the floor out from under everything below the distribution layer. the tool now depreciates in weeks. the operation can be copied in years. distribution is the only layer that gets harder to replicate the longer it exists, and when everything else is commoditizing, the last scarce thing tends to get priced as the asset.
the multiplier makes it manufacturable. a firm whose customers are themselves firms (a tax advisor serving 200 SMEs, an insurance broker sitting in front of dozens of policyholders) is one hop away from a large slice of an economy. fifteen hundred of those in a national market isn’t an audience in any normal sense. it’s closer to a sensor grid. and a sensor grid with a few years of calibration is the kind of thing insurers, banks, sovereign funds and macro desks already pay for. it’s the same buyer class that pays for a Bloomberg terminal or an ADP employment report. that buyer doesn’t care about your SaaS margins.
the reframe
i’d stop thinking of distribution as a line in the go-to-market budget and start thinking of it as the thing the go-to-market budget is building. the software depreciates. the operation replicates. the distribution, if you hold it and instrument it, is the only part that compounds.
every deal above is a buyer who worked that out before the people selling to them did. which leaves the uncomfortable question: are you building that asset, or building it for someone else?
this is the thesis behind the prediction economy, a six-chapter whitepaper on why distribution is the last moat in vertical software. published at predictioneconomy.ai