The Wall Street Journal published a number last month that deserves more attention than it is going to get. At the current pace of exits, private equity would need roughly nine years to clear the companies it already owns. Not to deploy the next fund. Simply to sell what is already in the portfolio. About 13,500 US companies sat in buyout portfolios as of June 30, according to PitchBook data analyzed by PricewaterhouseCoopers. Nearly 4,000 have been held six years or longer. Roughly 1,500 have been held for nine. An industry that underwrites three to five year holds has, in aggregate, become a parking lot.
How the Backlog Formed
None of this happened suddenly. Firms bought aggressively through 2020 and 2021, when capital was cheap and entry multiples reflected it. Debt was sized against growth rates the remote work era made look permanent. Then rates rose, the financing math stopped working for the next buyer in the chain, and the IPO window narrowed to a crack. Holding periods stretched. Marks drifted. Distributions slowed, limited partners had less cash to recycle into new commitments, and that pressured everything downstream.
The industry’s response was rational and entirely human. Nobody wanted to sell at a loss. Waiting looked free, because holding is the only decision that stays reversible. A mark can be defended, revised, or eventually vindicated. A sale is permanent, and it prices every comparable asset still sitting in the portfolio. Raymond James’ private capital advisory group made the point plainly in the same article: the 2021 vintage assets are the hardest to move, precisely because of the distance between where valuations were and where they are. So, sponsors waited. The inventory grew.
The Twelve Hundred
Software is a small share of the count and a large share of the problem. Only about 1,200 of those 13,500 companies sit in the software sector. But they absorbed an outsize portion of the industry’s capital, bought at the multiples software commanded when software was the most reliable compounding asset on the board.
Then the ground moved. Public software comparables repriced sharply earlier this year on deepening concerns about what artificial intelligence does to the sector, and private markets cannot ignore public comps indefinitely. The debt taken on against pandemic era growth assumptions did not reprice. The growth did. What remains is a cohort of leveraged, mature SaaS businesses whose owners will not sell at the new number and cannot argue convincingly for the old one.
Here is what makes this different from an ordinary down cycle. A buyer evaluating a 2021 vintage SaaS asset today is not only asking whether the multiple is fair. They are asking a harder question underneath it. Is this codebase an asset or a liability in a market where AI is compressing both the cost of building software and the price of buying it? That question does not resolve with patience. It resolves with evidence, or it resolves against you.
What the Next Decade Looks Like for Pre-AI SaaS
There are four paths out, and none of them are free.
- You can wait. Waiting costs carry, costs LP patience, and costs optionality, and it rests on the assumption that the market eventually returns to your number. For an asset underwritten against a growth curve that may no longer exist, that assumption is coming under some intense scrutiny.
- You can move the asset into a continuation vehicle, increasingly the default move. But continuation vehicles do not make the valuation question disappear. They relocate it. Now you are defending the remaining value creation plan to new investors whose due diligence will harder this round, than last, because they are buying at today’s price with today’s information.
- You can sell at a discount. This is the path worth sitting with, because the discount is rarely arbitrary. It is priced off what the buyer’s technical diligence finds. Architectural drift. Accumulated technical debt. Undocumented dependencies nobody left at the company can explain. A codebase the buyer’s engineers cannot confidently extend, integrate, or point a modern AI toolchain at. Each finding arrives in week five with a number attached, and that number comes out of your proceeds.
- You can reposition the asset itself. Not as a SaaS product with AI features bolted to the side, but as software with agents native to the application, working inside the workflow, with real access to the data underneath it. That is a different category in a buyer’s mind, and it is priced differently. This path is the only path that changes the asset rather than the timing.
Why Bolted-On AI Fails
The problem is that most sponsors have already watched someone try the cheap version of this. An agent gets layered onto an application that was never structured to support one, and it fails. The industry has now run this experiment at scale. Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. MIT’s study of enterprise generative AI found that against 30 to 40 billion dollars of investment, roughly 95 percent of pilots produced no measurable impact on profit and loss. Not modest returns. Nothing. Zero.
The reason is not the model, and it is not the prompt. MIT located the failure in brittle workflows and misalignment with how the business actually runs. Gartner is more direct: integrating agents into legacy systems is technically complex, disrupts workflows, and demands costly modification, which is why it recommends rebuilding the workflow rather than decorating it. An agent needs to reach into a system’s data, understand its business logic, and act through stable interfaces. Applications carrying years of accumulated technical debt offer none of that. The logic is buried and duplicated, the data is trapped behind undocumented pathways, and the interfaces are calibrated to assumptions nobody wrote down. The agent has nothing to hold onto. The relationship is direct and it is unforgiving: the higher the technical debt, the lower the probability that embedded AI works at all.
Which means AI enablement is not a feature decision. It is an architectural one, and it has a prerequisite. You cannot skip the remediation, because the thing that blocks the agent is the same thing the buyer is going to discount you for. Fix it once, and you have addressed both.
The next decade for pre-AI SaaS will be decided by whether the asset’s technical foundation can carry an AI era story. Not by how fast it once grew in a market that has since moved on.
The Part That Is Actually Within Your Control
You cannot control rates. You cannot reopen the IPO window, and you cannot talk the public or private markets out of how they are pricing software. Those are conditions. They will do what they do, with no regard for your fund’s vintage year.
What you can control is what a buyer finds when they open the hood.
What a buyer finds is not an opinion, and it is not a negotiating posture. It is a set of specific, measurable, and largely remediable technical findings. Fixing them is exit value defense, and it is worth doing. But defense only protects the number you already have. The fourth path is where the interesting math lives.
Consider what a buyer is actually pricing. Traditional SaaS is being marked against a comp set the market has decided is structurally exposed to AI. The SEG SaaS Index fell roughly 25 percent in the first quarter of 2026, and SEG attributes part of that compression directly to AI revenue uncertainty. Software with native agents is being marked against a different comp set entirely. SEG Research documents a one to three times multiple premium for AI-native software over comparable non-AI peers, and roughly eight in ten private equity and strategic buyers report they are already paying it in closed transactions. Eighty-seven percent expect that premium to hold or grow. The risk runs the other way as well. One in five strategic buyers walked away from a deal last year over AI exposure concerns, which means the question is not always what discount you accept. Sometimes it is whether a bid arrives at all.
Two-thirds of buyers say the companies they are evaluating show only limited AI adoption. The demand is priced. The supply is not there. What separates the two is almost never ambition, and it is almost always the codebase. Repositioning the asset does not merely slow the erosion. It moves the company out of the comp set that is compressing and into the one that is expanding. That is what multiple expansion looks like when it is engineered rather than hoped for.
It also changes who is in the room. A leveraged, aging SaaS platform draws secondary buyouts and opportunistic bids from sponsors who know you are running out of clock. An application with agents working natively inside the workflow, with clean access to its own data, draws strategics and AI-forward acquirers who are buying capability rather than cash flow. More bidders, better bidders, and a story that does not depend on the IPO window reopening on your schedule.
The prerequisite is unchanged. You cannot get there on a codebase that will not support it, and you cannot know whether yours will until someone measures it. We give you diligence-grade evidence of a software asset’s technical health, measured rather than asserted. If the health of the software asset needs to be improved, we can do that also – fast and at low risk. Until then, the useful question is not whether your software assets carry technical debt. They do. It is whether you can put a number on it before a buyer does it for you.
Let’s Talk
If you are holding software assets you cannot yet exit at your number, the most valuable thing you can do this quarter is find out precisely what a buyer will find. Book a call with Aspen using the following Calendly link: https://calendly.com/aspen-ess/aspen-ess-discovery-chat, and we can look at it together and develop a plan to create an AI era application.
