The Citrini Miscalculation OR How I Learned Not to Fear Agentic AI and Sleep Better at Night
- Ariel Steinlauf
- Apr 28
- 6 min read
Updated: May 13
On a Sunday night in late February, a Substack essay landed in hedge fund managers’ inboxes like a letter from a future nobody wanted. By Monday morning, it had erased roughly $300 billion in market value. Blackstone dropped 6.2% in a single session. KKR fell nearly 9%. IBM logged its worst day since 2000. DataDog, CrowdStrike, and Zscaler each shed more than 9%.
The piece was "The 2028 Global Intelligence Crisis" by Citrini Research. The authors had the courtesy to frame it as a thought exercise. Markets did not return the favor.
Two months later, the impact has made its way to the private equity sector and has surfaced material questions – and potential solutions – both of which have been hiding in plain sight for a while.
It’s the software, stupid
The Citrini thesis, stripped down to its core, is a PE fund problem. Not a macro problem. Not a credit market problem. A find-a-business, fix-it, sell-it-at-a-markup problem.
Here's why. For most of the 2018–2024 window, the most popular answer to the question "What should we buy?" was “Software!” Sticky recurring revenue, high margins, predictable cash flows. What’s not to like? PE sponsors paid 20–30x EBITDA for the pleasure of owning these businesses. The entire value creation thesis rested on one assumption: that the ARR would stay, well, recurring. The problem is that, much like in Angry Birds, agentic AI just hit the one block holding the fortress together, and everything came tumbling down. The pigs (no disrespect to the software companies) didn't see it coming.
When a Fortune 500 procurement manager can credibly threaten to rebuild a $500,000 SaaS contract in-house using AI agents — and some of them can — the renewal dynamic changes entirely. "ARR" turns out to have been a behavioral assumption masquerading as a financial one. Customers renewed because switching costs were high and the process painful, not because the software was irreplaceable. Once it became replaceable, the moat instantly dried out. And that feeds Citrini's loop: AI disrupts the software thesis, sponsors respond by cutting costs and investing more in AI capability, the software gets more disrupted, the company's enterprise value erodes fast. A flywheel that ran beautifully in one direction is now running in reverse.
Public markets moved first, as they always do. Software stocks shed nearly 30% between October 2025 and February 2026. The message was clear: the market was repricing exposure to businesses that PE had spent a decade financing.
Private credit became the first private market indicator to crack — and boy, did it crack. $130 billion in software acquisition-related loans were trading below 90 cents on the dollar. The canary, to use Citrini’s metaphor, was already on the floor.
The contagion then spread to PE. When private credit tightens on software, LBO financing gets more cautious across the board — not just for pure SaaS, but for any business with software characteristics the lender can't underwrite with confidence. Exit windows narrow. Sponsor-to-sponsor sale processes slow. And the large platform firms that serve as primary exit buyers for mid-market assets are now managing impairments in their own portfolios rather than hunting for acquisitions. The ripple hits every GP with a hold period expiring on a business that even faintly resembles the disruption thesis — regardless of whether their fund has a dollar of software exposure.
That's the reckoning Citrini really triggered. Not a credit event, but a portfolio one. PE firms that bought well are now asking whether they can still sell — and at what price, and to whom, and when.
Indeed, Citrini identified a real structural vulnerability. But did they identify the right outcome? That's where the miscalculation begins.
The Physical Economy Doesn't Care About Your Feelings (Or About AI)
The part Citrini missed entirely is that the same AI buildout that's destroying software pricing power is generating structural, decade-long demand for physical-world businesses that no one was paying attention to.
Data centers are projected to consume 6-8% of US electricity by 2030 (and that was before even accounting for AI), up from under 2% today. That growth requires power generation, grid infrastructure, real estate, cooling systems, and construction at a scale that dwarfs the software layer sitting on top of it. Stonepeak — focused on infrastructure and energy — didn't build one of the most competitive alternative asset platforms in the market by accident. They saw the physical-world thesis before it was obvious. Firms investing in healthcare services — home health, behavioral health, physical therapy — are insulated from AI displacement in ways that no enterprise SaaS company is. These businesses touch patients. They require licensed professionals in rooms with other humans. For now, the AI agent can't do that — and 'for now' may be the most important qualifier in this entire conversation.
As Scott Galloway called it correctly, if not approvingly: the US is essentially a giant bet on AI. He's not wrong — but the bet isn't only placed on the software layer. It's placed on everything required to run the software layer, and on the services that the software can't replace.
For sector-focused lower middle-market and middle-market PE firms — the kind that have spent the last decade investing in industrials, healthcare services, and infrastructure while the large platforms chased software returns — this is their moment. They're not managing a crisis. They're managing a period of exceptional demand growth, with entry multiples that software-heavy competitors can't match because those competitors are distracted by impairments in their own portfolios.
Revenue Growth Is Not Zero
The assumption that makes Citrini's darkest scenario work — and that a surprising number of PE practitioners are quietly accepting without realizing it – is that in an AI-disrupted environment, value creation collapses to cost reduction. That cost cuts are the only thesis now.
It isn’t. And the market conditions are making that clearer by the quarter.
In a recent podcast conversation with Triton’s Peder Prahl, Hugh MacArthur summed up the old PE playbook plainly: the traditional value investing reflex was to assume the revenue line is zero, then make all the money by reengineering costs. That worked when cheap debt and multiple expansion could carry a deal. It became a value trap when those tailwinds died. Prahl — one of Europe's most disciplined value-oriented investors — reached the same conclusion the hard way, auditing his own portfolio failures before rebuilding Triton's entire investment framework around one non-negotiable: growing markets with growing profit pools. His operating sequence, put simply, is: buy well, fix and expand, sell growth. Not buy and cut. Sell growth.
The math demands it. To hit a 2.5x MOIC in today's deal environment requires roughly 12% annualized EBITDA growth — twice what the 2015-era playbook needed. Cost cuts alone don't get you there.
This is precisely what the Citrini framing misses: agentic AI doesn't threaten all revenue models equally. For physical, non-software businesses in growing markets, it actively builds them. To illustrate, consider an agribusiness platform that deploys AI-driven yield modeling to optimize its own operations discovers it now owns a precision advisory capability its regional grower network will pay for — internal tool becomes new revenue line. A specialty re/insurer whose legacy actuarial process made entering new risk categories prohibitively slow finds that AI underwriting compresses that timeline from years to months — new lines of business that previously weren't economical to pursue. A gaming IP platform that was running out of content runway uses AI-generated environments and dynamic quests to extend game lifecycle and create monetization events that didn't exist in the original product — no new development cycle required.
These are fix-and-expand stories powered by tools that didn't exist three years ago. The GPs who treat AI as a revenue growth instrument — not just a margin lever — are the ones who will generate returns the Citrini scenario declares unavailable.
The industry is adapting. The question is how fast, and which firms lead it.
Why I Sleep Well
The Citrini miscalculation wasn't in the diagnosis. The loop is real, the marks are real, and the SaaS LBO vintage of 2018–2023 is going to produce some genuinely uncomfortable LP conversations. Where Citrini's analysis went off was in modeling the disruption and ignoring the adaptation — assuming the industry would absorb the AI transition as a passive victim rather than as a market full of actors with $900 billion in dry powder, sector expertise, and a front-row seat to the most powerful operational toolkit in the history of the asset class.
The firms built around physical businesses in growing markets aren't managing a crisis right now. They're managing an opportunity that got considerably cheaper in February. The ones with the discipline to treat agentic AI as a revenue growth engine — not just a margin lever — are going to generate the returns that the Citrini scenario declares unavailable. That's not optimism. That's the math.
The firms that get this right are already visible if you know where to look. That’s one part of the story. Helping them execute is another. Both matter. That's why I sleep well.
