AI Is Being Bought, Not Built

In the twelve months to mid-2026, one procurement-software company made four acquisitions. Coupa — owned by Thoma Bravo — bought Cirtuo, Scoutbee, Rossum, and Tonkean, each one folding a piece of AI-driven capability into the same platform. None of it was built in-house. All of it was bought.

That cadence is no longer unusual. It is the playbook.

The shift

For a decade, the reflexive answer to "how do we add AI?" was to build it — hire the team, ship the feature, iterate. That answer is quietly being replaced. The scarce inputs to good AI — specialized talent, proprietary data, and a working product with real users — are hard to grow on a roadmap timeline and easy to acquire outright. So the sponsors who move fastest have stopped waiting. They are assembling AI the way private equity has always assembled everything else: through M&A.

This is the distinction that matters, and it is easy to miss. There is a large, noisy category of "PE firms that invest in AI" — funds writing checks into AI startups the way everyone writes checks into AI startups. That is not the interesting part. The interesting part is a smaller, more deliberate group using platform M&A, buy-and-build, and strategic tuck-ins to fold AI capability into a software or services platform they already control.

One group is placing bets. The other is building a machine.

The four shapes of an AI roll-up

Watch enough deals and the pattern resolves into a handful of recognizable shapes.

The most common is the software buy-and-build with AI tuck-ins — an established platform acquiring smaller AI companies to accelerate its roadmap. Coupa is the clean example: four AI acquisitions absorbed into one platform in a year.

The second is the platform assembled around data — a sponsor combining several assets into a single data-and-AI engine. New Mountain built a healthcare-technology platform out of three separate companies, then added the AI-native Machinify to sharpen it.

The third is the boldest: the sponsor-created, AI-native company, built from scratch to scale by acquisition. In 2026, Blackstone and Hellman & Friedman did not buy into an AI services firm — they created one, acquired Fractional AI as its operating core, and launched it as Ode in partnership with Anthropic.

The fourth is the underwritten consolidation thesis, where the roll-up is the reason for the deal. When Vector Capital acquired Bigtincan and combined it with Showpad, it said plainly it had underwritten a consolidation thesis in a fragmented market. The M&A was not a follow-on. It was the plan.

How to tell a real one from the noise

If you want to separate a genuine AI roll-up from a firm that merely owns some AI companies, look for two things together: a stated buy-and-build or consolidation thesis, and at least one completed AI acquisition or a real AI operating capability. Either signal on its own is common. Both at once are rare — and that is the tell. A firm with a fund built explicitly around software and artificial intelligence, or a dedicated in-house AI team wired into its platforms, is playing the game. A firm with a couple of AI logos in an otherwise generalist portfolio usually is not.

The buyer is often the portfolio company

There is a structural quirk worth understanding, especially if you are on the receiving end of this. The acquirer is frequently not the fund — it is the portfolio company. Coupa buys Tonkean. Datasite — backed by CapVest, with a $500 million mandate to expand its intelligence products organically and inorganically — buys Blueflame. The sponsor supplies the capital, the mandate, and the operating playbook, but the deal is executed one level down. Which means there is rarely a single buyer in these situations. There are usually three sets of hands on the wheel: the deal partner, the value-creation team, and the platform company's CEO or corp-dev lead.

What becomes scarce

Here is the part that should interest anyone building a company in AI's path. When capability is bought rather than built, the definition of a valuable target changes. Another thin wrapper on a foundation model is not scarce. What is scarce is the thing you cannot spin up in a quarter: proprietary data, genuine organic usage, a distribution wedge, and the capital efficiency that comes from having grown without burning. The assets that draw real attention in this cycle tend to be quietly excellent on exactly those axes — real users, real data, real retention — regardless of how loud their branding is.

Where it goes next

As the playbook matures, the constraint shifts from sourcing to proof. When a platform is acquiring AI capability every quarter, the diligence question stops being "what is the vision?" and becomes "show me the real usage, the real data rights, the real retention." The advantage on the sell side moves to whoever can answer that instantly and verifiably — not through a static deck assembled the week before, but with numbers that stand up to inspection. The era of taking the narrative on faith is closing.

The mental model is simple, even if the execution is not. In 2026, AI is being bought, not built. The firms that internalized that early are compounding platforms while everyone else is still debating build-versus-buy. And the companies worth buying are the ones that can prove, on demand, that what they have built is real.

The takeaway

• AI capability is increasingly bought, not built — assembled through M&A rather than grown on a roadmap.
• The real roll-ups take four shapes: software buy-and-build with AI tuck-ins, a platform assembled around data, a sponsor-created AI-native company, and an underwritten consolidation thesis.
• The tell of a genuine one: a stated buy-and-build thesis and a completed AI acquisition or real AI operating capability — not just AI logos in the portfolio.
• The acquirer is often the portfolio company, not the fund — so there are usually three buyers, not one.
• As the playbook matures, the scarce, valuable target is the one with proprietary data and real organic usage that can be verified on demand.