No Salvage Value

Jim Feldkamp · Managing Partner, Alary Capital · September 2026

Prefer as a PDF? Request it by email.

When a software loan turns there is no collateral to liquidate toward. Enterprise value is recurring revenue carried on confidence, and confidence does not sit still while a workout runs. This paper is about what that changes for a lender: why the software cohort financed at the 2021 peak is the hardest inventory in private credit, why its exit is not simply late, and what has to be underwritten before a process starts.

1. There is no salvage value underneath a software company

A lender to an industrial borrower recovers through things. Inventory can be counted and sold. Receivables can be collected. Machines and real estate carry a price at liquidation that is low but real, and that price is what makes patience rational: the asset sits still while the parties negotiate. Every convention of the workout playbook, wait for the covenant, appoint the advisor, run the process, was written for collateral that behaves this way.

A software company has none of it. Its enterprise value is recurring revenue, and recurring revenue is not an asset in the sense the playbook means. It is a set of renewal decisions taken by customers who can leave, executed by engineers who can resign, on a product that has to keep being built. Remove the company’s confidence and there is no residual underneath: no salvage value to liquidate toward and no floor to price against. Value does not decline in an orderly way. It evaporates.

That single difference reorders what a lender should do after a credit turns. In an asset-backed workout time is roughly neutral and often positive, because the collateral holds its value while the lender improves its position. In a software workout time is the variable that destroys the recovery: a quarter of drift costs renewals that will not come back, engineers who will not return, and a roadmap that falls a quarter further behind. The playbook’s founding assumption, that waiting preserves optionality, is not merely weaker in software. It is inverted.

The practical consequence is that the recovery question has to be answered earlier in software than in any asset-backed credit, because the thing being recovered is being consumed while the question stays open. What follows is why that question is now most pressing for one particular cohort, why the market that would answer it has closed around exactly those companies, and what a lender can underwrite before a process starts.

2. The reset, and the debt that came with it

Software’s valuation correction is the steepest of any sector, and it is not a dip below trend. At the 2021 peak the median public SaaS company traded at roughly 18x to 19x revenue. As of mid-2026 the public median is 3.2x trailing revenue, down from 5.7x a year earlier. Private acquisitions traveled the same road from a lower peak, from a median of 6.4x revenue in 2021 to roughly 3.1x now.¹

Read that as a return to the decade’s norm rather than a crash, because that is what it is, and it is worse news for the cohort than a crash would be. A crash implies a rebound. A normalization implies that the companies bought at the peak were bought at prices the market is not going to revisit, and that their owners hold assets whose marks depend on a multiple that is not returning on any schedule the loan can survive.

That vintage carries a second feature, quieter and more consequential for a lender. The landmark software take-privates of the period were financed with ARR-based unitranches: loans underwritten to recurring revenue rather than to earnings, on the theory that growth would produce the EBITDA later. Those structures were never meant to stay that way. They convert to conventional EBITDA-based covenants two to four years after closing.² Later is now.

So the cohort meets both problems in the same year. A company that did not grow into its multiple is also a company that did not grow into its covenants, and it discovers the second at the moment the first makes a sale unattractive to its owner. That is the shape of the inventory sitting in private credit books today. Not a scattering of troubled names, but a vintage with a common structural defect arriving on a common schedule.

3. The lenders holding it, and the amendment wave

Software is the largest single sector exposure in direct lending. The Bank for International Settlements counts more than $500B of loans to SaaS companies, about 19% of all direct loans, and the Boston Fed puts internet and software near 20% of the median BDC portfolio, the largest sector exposure in the book, with some lenders above a third.³ A Wall Street Journal analysis argued the true figure at several large funds is closer to a quarter once borrowers reclassified into other industries are counted.⁴ However the line is drawn, the conclusion holds: the cohort described above is not a niche inside private credit. It is close to the center of it.

And the credit is deteriorating in the aggregate data. Lincoln International’s valuation work shows 11% of private credit loans paid interest in kind during 2025, up from 7% in 2021, and that 56% of that PIK was added after closing rather than written into the original terms.⁵ The distinction carries the signal: PIK negotiated at origination is a structure, while PIK added later is a borrower that stopped being able to pay cash. Covenant defaults ran about 3% a quarter and lenders processed more than 875 amendments in a year.⁵ Fitch’s US private credit default rate reached a record 6.0% for the twelve months through June.⁶

When a software LBO cannot refinance at par and cannot sell at its mark, the observable event is an amendment, and the wave is now public rather than private.

The first four are the same sentence written in loan documents: the exit did not come, so the debt must wait. Medallia is what happens when the waiting stops. All five were decisions about the recovery, and each was taken without the one analysis that would have priced it.

4. There are buyers. They are not buying this.

It would be wrong to read any of the above as a closed market. Software deal count is at a record: the twelve months through June saw 2,784 SaaS transactions, most of them small and vertically focused add-ons.¹ Mega take-privates are back, with Electronic Arts at $55B and Dayforce at $12.3B, and reports in mid-August of Silver Lake in talks to take Workday private.¹²

Look at what actually clears and the pattern is unmistakable. Large, profitable, moderately priced assets clear. Small tuck-ins clear. The middle is empty. US software platform buyouts are tracking toward a decade low, roughly $39B annualized against $156B in 2025, and tech buyout volume fell 78% year over year in the first half.¹³

The middle is precisely where a credit-stressed, middle-market software company sits. It is too small to be one of the handful of assets a strategic buyer or a mega-fund will underwrite through a difficult story, and too large to be absorbed as an add-on by a platform that is under pressure itself. One software banker described sponsors’ posture toward the sector as “a bit like getting punched in the gut. You lose your breath for a period of time and are cautious about going back.”¹³

Over all of it hangs the AI question. Bain measured an 8% decline in software valuations in the first quarter of 2026 alone and named AI disruption one leg of the year’s “triple shock.”¹⁴ Vista has publicly moved its benchmark from the Rule of 40 to a “Rule of 60.”¹⁵ Buyers now screen valuation, durability of growth and AI exposure before granting a management meeting. None of this means a given company cannot be sold. It means the burden of proof has moved decisively to the seller, and that the proof has to survive diligence rather than merely open it.

5. Where these names sit in the queue

All of this happens inside a market that has largely stopped clearing. PitchBook’s mid-year data put 13,509 sponsor-owned companies in the United States, part of a global count above 33,000.¹⁶ Against a US exit pace of roughly 1,600 companies a year, that is more than eight years of inventory at the current rate, by PitchBook’s own arithmetic.¹⁷ The queue is measured in years rather than quarters, and it lengthened again in the second quarter, when US exit value roughly halved from the first.¹⁶

It is also aging, and aging unevenly. The median US company exited this year had been held 6.4 years, and the share of portfolio companies held five years or longer is at its highest in nearly a decade.¹⁶¹⁸ FTI Consulting counts roughly 1,400 investments above the historical norm, about 11% of US inventory, and finds retention of the 2021 and 2022 cohorts higher than any vintage since 2004. Those large buyouts were done at a record 13x EBITDA against a historical band of 9x to 11x.¹⁹ The companies least likely to be sold are the ones bought at the top, which is the cohort this paper has been describing.

Every route out narrowed at the same time. Since 2022 there have been 70 sponsor-backed IPOs on US exchanges against 424 in the five years before, and that route was never open to a credit-stressed middle-market company in any case.²⁰

The sponsor-to-sponsor trade, the workhorse of the last decade, fell 57% in the second quarter to $24.5B across 94 deals, the lowest quarterly count in at least a decade, while sales to corporate acquirers fell 63%.²¹ What grew instead was engineered liquidity: a record $120B of secondaries in the first half, with continuation funds at 86% of GP-led deal count,²² and an estimated $70B of NAV lending in 2025.²³ These instruments have real uses, but for a stressed company they defer the sale rather than complete it. A continuation fund does not establish whether the company can be sold. It postpones the date on which someone finds out, and the claim ages while it waits.

The sponsors say the quiet part in earnings language. Apollo, reporting principal investing income of $16M in the second quarter against $75M in the first, told investors asset sales had been “prudently delayed” while conditions are “less accommodative for monetization activity.”²⁴ Delay is now the reported strategy. What it costs, and who pays for it, is the subject of the next section.

6. What a year of waiting costs

Waiting feels free because nothing happens on the day it is chosen. It is not free, and in software it is expensive in a specific and traceable way.

Follow the mechanics of one more amendment. The amendment converts cash interest to PIK. The PIK compounds above the equity. The compounding raises the payoff any eventual sale must clear. The higher hurdle narrows the field of buyers able to meet it. The narrower field weakens the process the amendment was taken to protect. The wait raises the very bar it was taken to avoid.

Then add what the year does to the company underneath. Another twelve months of renewals decided by customers watching a product whose roadmap is frozen. Another twelve months for the employees a buyer would have wanted to keep to be recruited by companies that are not in a workout. Another audit cycle. And another year for the AI question to harden from a diligence topic into a discount, which is the difference between a buyer asking how the company will respond and a buyer pricing the assumption that it will not. The mark stays flat. The company underneath it does not.

Everyone at the table has a private reason to let the quarter pass. Sponsors defer because a sale crystallizes the mark. Lenders defer because an amendment defers the loss. Bankers rarely argue for waiting, but they are paid on the attempt rather than the outcome, so the corrective conversation, whether this company can be sold at all and at what, tends not to happen anywhere. The company pays for all of it.

For the lender the cost is now on a quarterly clock. Watchlist and nonaccrual disclosures at listed funds are public and at multi-year highs in the aggregate, and in the second quarter investors asked private credit funds for $15.6B back and received $5.9B.²⁵ Every quarter a stressed name stays unresolved it migrates toward the disclosure that invites the hardest questions. The choice the market actually offers is not between selling now and selling later at a better price. It is between running the recovery on evidence and a timetable, and holding with no thesis except hope for a different market.

7. What kills a software process in diligence

Processes die in diligence, and they are dying there more often. In Axial’s study of broken lower-middle-market deals, 46.6% of collapsed letters of intent died on diligence findings, and the share killed specifically by quality-of-earnings discrepancies doubled between 2023 and 2025.²⁶ Two-thirds of GPs told Investec they had seen more broken processes than in prior years,²⁷ and nearly three-quarters of senior M&A executives expect diligence to become more complex still.²⁸

For a software target the examination surface is unusually wide, and every part of it is a place where a seller’s own numbers can fail to survive contact. Revenue recognition, because ratable revenue can be presented several defensible ways and a buyer will test all of them. Net revenue retention, because the seller’s cohort definitions rarely match the buyer’s. Customer concentration and the renewal calendar, because a buyer prices the next twelve months of renewals rather than the last twelve. Security posture, because a single unremediated finding reprices the deal. And now AI resilience, because no buyer will fund a product whose core function it expects to be commoditized.

A defect the seller could have found in month one surfaces in month five, at maximum leverage for the buyer and maximum fatigue for everyone else. That is why the first process is the asset. A second is different in kind, because it opens with the buyer’s first question, why is this back? Bankers talk, buyers compare notes, and the company re-enters the market wearing the failure. No index publishes what that costs. It shows up in who declines to bid, in the retrade, and in financing that gets harder. For the lender the translation is exact: a failed process is a failed recovery, and the discount that follows comes out of the claim.

8. Underwrite the recovery like the loan

Everything above converges on one asymmetry. The loan on the fund’s book was underwritten: modeled, stress-tested, papered, approved by a committee that could say no. The recovery that now determines that loan’s outcome is, at most firms, not underwritten at all. A path is chosen, forbear, extend, operate or sell, on an advisor’s cash flow, a banker’s market read and a fee structure that pays on the attempt.

Underwriting the recovery starts with the objective, because best recovery is not one number. On one loan it is speed, out by quarter end and flexible on the mark. On another it is maximum value, with the patience to earn it. On most it is the honest question of which of the two the credit can support. The lender sets the objective and the underwriting prices the paths against it.

In a software credit that means four straight answers, evidenced before anyone goes to market. What the company is actually worth, on the multiples clearing now rather than the multiple it was bought at. Who actually buys it, by name, given that the middle of the market is where buyers are thinnest. What has to be fixed before a buyer’s diligence finds it, across the surface described above. And how long that takes, measured against the renewal calendar and the covenant conversion date rather than against a banker’s calendar.

It means the fix list: every defect a buyer’s diligence would surface, priced in enterprise-value terms, ordered by what each is worth at exit, and worked down against a committed launch date. In software that list is rarely financial housekeeping. It is the cohort file that will not reconcile, the top-ten customer whose contract has not been renewed, the security finding nobody owns, and the AI answer a management team has never had to give under examination.

It means a two-scenario recovery analysis, what a sale supports prepared against what it supports as-is, so speed and value are priced against each other rather than argued about. And sometimes it means the no-go call: the finding that the company as it stands should not go to market, together with exactly what would change the answer and what recovery options remain. Given how little sits underneath a software company at liquidation, the no-go call is the most valuable output the discipline produces, and the least likely to be reached by anyone paid on the attempt.

None of this is process for its own sake. It is how the first process is protected: the defects are found by the seller’s own diligence instead of the buyer’s, the buyer list is real before the launch decision is made, and the process, when it launches, is built to close on the first trip. The smooth sale, the one that closes on schedule at a defensible price, is not a market outcome. It is a prepared one, and for the lender it is the recovery itself.

9. About Alary Capital

Alary Capital works with lenders to get the best recovery from challenged asset-light credits, with operating capability, underwriting discipline and our own capital. Software and tech-enabled services are where we started and where the argument above is sharpest, but the same underwriting applies wherever enterprise value rests on recurring revenue rather than on assets. We work as an investor that bids or as an operating partner with no bid, and the lender chooses which seat we take on a company, in writing, before the work starts.

You set the recovery objective. We underwrite the path to it.

Jim Feldkamp, Managing Partner, Alary Capital · info@alarycapital.com

Sources and notes

Market data as of August 17, 2026, except the software platform-buyout and valuation series, which are as reported through August 2026. Figures are aggregate, publicly reported data; where a figure is an estimate or a press-reported computation, the note says so.

Talk to us.