By the end of 2025, 84% of private equity firms had appointed a chief AI officer, according to EY's Q4 2025 report on AI in private equity.

Why the rush? Buyout returns got much harder to hit, and the only place left to find them is inside the companies themselves. Better margins, faster growth, lower cost to serve. AI is the biggest lever anyone has for all three, so that's where the money went.

Hashi's take: most people assume the play here is cost-cutting. That's not wrong — cutting cost is one of the ways PE has always made a company worth more. But the bigger prize shows up at the portfolio company level: pay to learn AI once, then multiply it across dozens of businesses. No other kind of owner can do that.

If someone on your board still thinks the old math applies, forward this their way.

Bain's 2026 Global Private Equity Report has the cleanest version of this. Through the 2010s, a buyout needed about 5% annual EBITDA growth to return two and a half times the money over a five-year hold. Today the same deal needs ten to twelve percent. They call it "12 is the new 5."

Nothing about what limited partners expect changed. The cost of borrowing changed, and every deal doubled the amount of operating improvement it needs to clear the same bar.

Exits are recovering, and strongly — buyout exit value rose 47% in 2025 to $717 billion, the second-best year on record. It still isn't enough. Bain counts 32,000 unsold companies worth $3.8 trillion, and the cash going back to investors is running at a rate last seen in 2008, the fourth straight year below the long-run average.

That's a lot of businesses that have to be worth more before anyone can sell them.

So the industry moved where it looks for returns. In S&P Global's February survey, 72% now name operational improvement as their top value-creation lever, and 60% of general partners say higher capital costs are what forced the shift. You cannot financial-engineer twelve percent a year. It has to come out of the business.

AI and automation are what promise those extra points. That's the catalyst for the hiring.

Most companies adopting AI pay for the learning once and use it once. They hire the people, make the mistakes, find out which vendors oversell, work out what the governance actually has to say, discover which three use cases were never going to work. Every bit of that expense lands on one P&L.

A sponsor with eighty portfolio companies pays for that learning once and uses it eighty times.

That edge belongs to private equity specifically. Adams Street's review of how buyout managers operate describes the machinery already being built for it: centralised AI labs, cross-portfolio summits, curated supplier lists, technical advisors placed onto boards, shared playbooks designed to move a company from experiment to impact without repeating the discovery from scratch.

A mid-market company can't afford to learn AI properly. Eighty of them under one owner can.

And if it's set up right, it compounds. What one company works out flows up to the fund and sideways to the rest. The healthcare services business solves document intake, and the industrial distributor gets it in a quarter instead of a year. Structure the AI office properly and you don't have eighty separate experiments. You have one learning engine.

And the logic isn't exclusive to private equity. A holding company, a family office, a franchise group — anything with more than one business under one owner has a smaller version of the same edge. Private equity just owns the most of them, which is why the advantage is biggest there.

That's the prize. So how much of it actually exists today?

Two places to look: what the funds built for themselves, and what's running inside the companies they own.

The funds. Mergermarket reported in July that the Swedish buyout firm EQT has moved Motherbrain — an AI platform it has been building in-house for years — from finding deals to acting as a sounding board for its investment committee. The American firm Advent International trained a system on thirteen years of its own investment memos, including the deals it walked away from, so it can challenge the thinking behind every new one. Bain Capital built an "AI twin" of itself: a digital model of how the firm actually works, used to find where AI could redesign a job rather than just speed up the existing one.

All of it is the same structure: build the capability once, then use it on every deal that follows. And all of it is in one place. Sourcing, diligence, memos. The fund's own work.

The companies. FTI surveyed 200 fund and operating leaders in May and found 7% of portfolio companies have AI running at enterprise scale. In the same survey, 95% of funds said their AI initiatives met or beat the original business case.

FTI doesn't explain why those two numbers sit so far apart. What they do say is that the shift from experimentation to production "remains inconsistent across PortCos," and that "the gap between isolated success and enterprise-scale advantage is still wide."

So plenty of individual projects are working. Very few have reached across a whole company. And nobody is measuring the thing I most want to know — whether what one company worked out ever reaches the others.

Learning doesn't move between companies on its own. Someone has to carry it — notice what worked at one business, judge whether it applies to another, and get it installed there.

That's the job the industry has been hiring for. The 84% bought a structure: an AI office at the fund, run by a chief AI officer. Attached to it is a newer seat with a name of its own — the AI operating partner.

Korn Ferry describes what that job looks like in practice. Building automation playbooks for finance and accounting that take out roughly a quarter of the effort. Pushing AI into marketing, customer service, inventory, warehousing, legal. Embedding it in the product itself — their example is a healthcare platform using AI to match patients with caregivers. One person, or a small team, working across a dozen or more companies at once.

That is the learning engine, described as a job spec. And Heidrick & Struggles reports "an enormous spike" in hiring for it.

So the seat is real and the demand is real. The question is who gets hired into it.

Heidrick lists three sources of candidates: people who have done AI at a PE fund, consultants from Tier 1 firms with AI labs, and machine learning leads out of large financial services firms. Korn Ferry's version of the same list is wider, and includes entrepreneurs who built AI-driven companies and technology executives who have run large deployments.

One list is drawn almost entirely from finance. The other includes people who have changed how a company actually works. The engine needs the second kind, because the work happens on an operating floor and not in a memo.

Korn Ferry also names the failure mode. When the role is loosely defined, portfolio company leadership ends up managing yet another stakeholder from the fund, and the result is "uncontrolled pressure to adopt AI without business justification." That's what a badly built AI office produces — not compounding, just noise arriving from head office.

Both surveys agree on what is holding this back, and it isn't technology. FTI puts talent at 35%, the primary constraint on scaling. S&P puts lack of expertise at 49%, ahead of data privacy and model accuracy. Not compute. Not budget, which is enormous and still growing. People — specifically, someone who can hold a value creation plan in one hand and an operating floor in the other, and translate in both directions.

If you're running a portfolio company, that's your leverage. The fund needs that translation more than it needs another tool.

Firms that get this right — the right hires, pointed at the portfolio rather than at the fund — will pull real returns out of assets that look stuck today. Some of those 32,000 companies get sold because of it. Get it wrong and all you've bought is an expensive job title and more pressure on people who are already busy.

If you don't work in private equity, watch these firms anyway. Whatever they learn, they learn across dozens of businesses at once. Almost nobody else gets that vantage point, and those lessons travel.

One more, and it's speculation. The firms that mature at this shouldn't stop at their own portfolio. Build a machine that has genuinely transformed forty businesses and you're holding something every other company would pay for. AWS started as infrastructure Amazon built for itself.

I haven't spent a career in private equity and won't write as though I have. I've spent years on the unglamorous half of this — getting AI to work inside ordinary businesses run by people who don't call themselves technologists.

Keep reading!

- Hashi