BOTTOM LINE UPFRONT
PE has poured $15.5 billion into AI deployment engines this year, yet most of that spend can’t prove it moved a single EBITDA lever. The engineers are building. The people who know what a PE-backed CFO actually needs aren’t in the room. Until that changes, the tail keeps wagging the dog.
If you shut down every AI initiative across your portfolio tomorrow, how many companies would miss their numbers? Across private capital markets, the answer is much smaller than anyone wants to admit. Ask that question in a board meeting and watch the room go quiet. Now look at what happened anyway.
Last month, OpenAI launched The Deployment Company alongside TPG, Bain Capital, Advent, and Brookfield, raising more than $4 billion for a venture valued at $10 billion. Within minutes, Anthropic announced its own $1.5 billion partnership with Blackstone, Hellman & Friedman, and Goldman Sachs to build a competing platform. Fifteen and a half billion dollars – OpenAI’s venture at $14 billion, Anthropic’s at $1.5 billion – walked into a room where engineers outnumber everyone who has ever closed a quarter and handed them the keys.
Every board wants an AI update, and every management team feels the pressure. PE has gotten very good at producing both, but not so good at producing EBITDA from any of it.
Let’s be clear: Buyers are indifferent to pilots. They pay for revenue growth and margin expansion and businesses that operate leaner than they did twelve months ago. PE has done this before, with ERPs, with data platforms, with digital transformation. Each time, genuine technology met serious capital and deployment became the objective while the business problem got treated as an implementation detail.
None of this is an argument against AI. AI will reshape PE-backed businesses in ways that are genuinely hard to overstate. It is an argument about who should be in the room before the engineers start building.
Spoiler alert: the right people aren’t there. And that’s because the incentive structure around AI is uniquely screwed up. In prior cycles, someone eventually got held accountable for whether it worked. AI has no correction mechanism. Nobody gets praised for killing a project and nobody gets promoted for saying the use case was not there.
The rewards flow entirely toward whoever launched something, announced something, showed momentum. All of which makes AI the first technology wave where deployment itself is the KPI. And it’s working in the worst possible way. Eighty-three percent of PE-backed companies have at least one AI pilot running. Only 18% are genuinely tied to value creation levers or pressure-tested against the questions a future buyer will ask in diligence. The rest is AI theater, and the people running it know it’s theater. But they’re doing it anyway because the optics of doing nothing are worse than the reality of doing the wrong thing.
The market is starting to ask harder questions. At SuperReturn Berlin this year, token consumption came up repeatedly, but the lens has changed. A year ago, high usage was evidence of progress. Today it is a cost line that needs to justify itself. The conversation has moved, and the tell is always the same: who was in the room, and who wasn’t.
The FDE was there, so somebody built something they thought was cool. The product manager who understands value creation was not there, so nobody defined the right problem first. The business analyst who could connect the solution to EBITDA was not there, so nobody asked if it was worth building. The frontline leader who could have said in five minutes whether anyone was going to change their behavior for this to work was not there, so the answer to that question arrived six months after launch. The thing got built, it demoed beautifully, and it moved nothing.
The firms arriving with the best intentions and engineering-first mandates are walking into portfolios and setting up this exact room. Real engineering firepower, almost no institutional feel for what a PE-backed CFO needs at month 18 versus month 48 of a hold period. The traditional advisory firms understand the business and are frantically hiring engineers to prove they can build. None of them have fixed the room.
The history of electrification is worth remembering here. When electricity arrived, factories swapped steam engines for electric motors and productivity barely moved. Companies had the technology and kept the same workflows, the same layouts, and the same operating model. The productivity gains came years later, when manufacturers redesigned factories around what electricity made possible. Most organizations are in that early phase with AI right now. The technology is there but the workflows have not caught up. The organizations that will pull ahead are the ones willing to redesign around what AI makes possible rather than layer it onto what already exists.
The firms that are already winning did not wait for the analogy to land. The business problem comes first, scoped by people who understand what moves a PE-backed business, mapped to a specific EBITDA lever, stress-tested against what a buyer will want to see in two years. Then the FDE builds.
Fifteen and a half billion dollars is currently doing it in the opposite order, and the bill is coming. Buyers are already asking about AI in diligence as a line item, and 41% of PE firms scaling AI across multiple portfolio companies have no operational playbook behind it. The post-mortems will say the technology worked and the deployment failed. CFOs will pay for that inevitable failure twice: once for the lesson and once for the outcome. Because the FDE model only works when someone in the room knows which problem is worth solving.
PE invented the concept of operational discipline. For thirty years, the industry has held management teams accountable for every dollar that didn’t move EBITDA. The AI era is no different. At some point the dog remembers it’s the dog.