The private equity playbook that drove fifteen years of TMT returns just lost all three of its engines at once. The firms willing to rebuild the thesis from scratch will own the next cycle. The ones still running the old play are walking into a wall in 2028.

The old play is dead

For most of the last fifteen years, private equity in technology, media, and telecom ran one reliable play. Buy a business with sticky, recurring revenue. Finance it with cheap, abundant debt. Improve the operations or roll up a fragmented niche. Then ride an ever-expanding multiple to a clean exit. Recurring software revenue made the model bankable, near-zero rates made it cheap, and a decade of rising valuations meant the multiple did most of the work on its own.

All three conditions reversed at once. Average software valuations fell from above 9x NTM revenue in early 2024 to about 4.8x by April 2026 [1]. Analysts expect the compression to hold and even continue hurting anything AI can touch [2]. Credit tightened right alongside it. Lenders want bigger equity cushions, loan-to-value sits near 50%-60%, and software leveraged loans now price around 755 basis points against 435 for everything else [3]. The asset class PE leaned on hardest is the one lenders are punishing most.

The bill comes due in 2028, when roughly $417 billion of software leveraged credit matures, most of it lower-rated [4]. Two of the playbook's three engines are already gone: cheap leverage and automatic multiple expansion. Exits are stalling too, because buyers and sellers can't agree on what a company is worth in an AI-disrupted market [2].

AI split the market in two

The second shift is what makes the first one permanent. AI has moved from narrative to measurable, and it splits software cleanly into winners and losers. Two questions decide which side a company lands. Can AI rebuild its moat? And does AI grow or shrink demand for what it sells?

On the defensible side are the businesses AI makes more valuable. Vertical software with proprietary data and complex, mission-critical workflows a general model can't reproduce. Identity and security platforms, because every new AI agent is a non-human user that must be governed. Infrastructure like Cloudflare and Snowflake, which only get busier as AI adoption climbs. And the high-trust, human-in-the-loop incumbents in regulated work, where a mistake is expensive [5].

Then there are the businesses AI hollows out. Seat-based software priced per human dies as a single agent does the work of several people. Labor-arbitrage IT and business-process firms billed by headcount are selling the exact effort AI now automates. Shallow point tools with no proprietary data and nothing to switch away from are cheap to rebuild in-house [5].

NRR is the cleanest read on stickiness, and it's sliding for the exposed names. ZoomInfo is stuck at 90% and walking away from per-seat pricing on purpose. Sylogist dropped from 108% to 98% in a single year. Atlassian, meanwhile, just booked its biggest quarter ever of competitive displacements as customers tore out legacy tools [6]. Scarce AI and security assets trade at wild multiples. Wiz changing hands near 64x revenue and CyberArk near 21x, while tired platforms go private at 2-8x [1].

Media tells the same story in a different language

Media is undergoing a similar transition. The business pivoted from "subscriber growth at any cost" to monetization that pays. The money is flowing toward advertising and connected TV, a market on track to roughly double from $44 billion in 2025 to $81 billion by 2030, overtaking linear television, with Google, Amazon, and Netflix set to take half [7]. The metric that matters now is ARPU, and how many subscribers are new versus trading down to cheaper ad tiers. The rest is draining away: growth without profit, overpriced output deals, fading linear advertising, and small streamers with no shot at ad-dollars.

AI is cutting production costs by an estimated 30%-50% and lifting margins, risking the value of all content libraries. Companies are accelerating, stretching, and revaluing their content libraries, which inflates the value without the business underneath changing. So, the margin math itself is now part of diligence.

What a defensible deal looks like now

Returns must close on operating improvement and debt paydown by themselves, without relying on an exit multiple. The moat needs to be the kind AI strengthens: proprietary data, embedded workflow, identity, infrastructure, regulated trust. Price the entry in case high rates, multiple compression, and AI pressure on revenue all land at once. Line up who buys at exit, because a re-rating won't rescue the deal.

These are strategy questions, answered before signing, not after. Start with the return lever the thesis leans on and ask if it still works in this regime. Then, question the moat against AI. Last, does the price still hold if rates, multiples, and AI all break together? Answering those strategic questions requires independent sector work, honest downside underwriting, and a read on the asset's real competitive position.

The capital is there. Deals are closing. The thesis hasn't caught up. Firms that rebuild and treat 2026 as a structural reset instead of a soft patch are the ones that will compound when the 2028 wall lands.