On August 4, 2026, Bending Spoons agreed to buy Airtable. The headlines said $2.25 billion. They said 80% off the 2021 peak. Both numbers are too generous.
The enterprise value is $1.285 billion.1 The $2.25 billion figure only appears after you add back Airtable's own net cash — $965 million of it.2 Airtable raised more than $1.4 billion over its life.3 So roughly two-thirds of this "exit" is investors receiving back money they handed over and the company never spent.
The operating business cleared at $1.285 billion against a peak valuation of $11.7 billion.2
That is 89% off.
The multiple fell harder than the valuation
Airtable's ARR is approximately $480 million as of June 2026, growing more than 20% year over year. Five hundred thousand organizations. Most of the Fortune 500.4
Enterprise value divided by ARR: 2.7x.
Nothing broke. Revenue went up. Customers stayed. The company is larger, more embedded, and better positioned than it was at the peak.
The valuation fell 80%. The multiple fell roughly 90% while the business grew.
Every dollar of the decline is repricing. None of it is deterioration.
A buyer in December 2021 paid $11.7 billion for a company that four and a half years later would do $480 million. That price was roughly 24x revenue the company had not yet earned. The same asset now clears at 2.7x revenue it has.
This is not a story about a company that failed. It is a story about what a working software company is worth when the multiple regime changes underneath it.
The secondary market was wrong by 44%
Here is the part that should alarm anyone holding private marks.
Airtable shares traded on secondary markets at a $4 billion valuation earlier this year.3 That was the skeptical price. The arms-length price. The one set by professional buyers with no stake in the founder's narrative.
The actual print is $2.25 billion equity, $1.285 billion enterprise.
The secondary market — the mechanism we rely on to discipline stale marks — was still roughly 44% above the clearing price. And that comparison uses the generous equity figure. Against enterprise value, it was off by a factor of three.
There is an alternative reading: Bending Spoons extracted a motivated-seller price, and the secondary marks were closer to fair value than the print suggests. The reading fails on one fact. This was a negotiated sale with both boards approving, and nobody outbid. If $4 billion was the fair number, $2.3 billion left a $1.7 billion arbitrage sitting on the table for any strategic, any sponsor, any secondary buyer with conviction. None appeared.
Jefferies reported that in 2025, tail-end private funds more than ten years old traded at 73 cents on the dollar of reported NAV, while funds under five years old cleared at 95 cents.5
Those now look like early readings, not floors.
What this actually is
It is a comp.
Private marks are not observations. They are estimates, produced by models, defended by the people whose fees depend on them. They hold until a transaction forces a number into the open.
Airtable just forced a number into the open.
Marks are opinions until somebody writes a check. Then they are prices, and prices propagate.
PitchBook's read was blunt: negative for private-market marks, positive for liquidity.6 That is the entire mechanism in nine words. Liquidity arrives, and the cost of it is that everyone else has to look at what liquidity costs.
Now count what stands behind it
Bain's 2026 Global Private Equity Report puts the exit backlog at roughly 32,000 unsold companies worth $3.8 trillion — up from about 29,000 companies at $3.6 trillion a year earlier, and $3.2 trillion at the end of 2023.7 Over $3 trillion of that is unrealized value inside global buyout portfolios.8
Distributions as a share of NAV have flatlined at 14%. That is a level last seen during the financial crisis.8
Average hold periods now run close to seven years.9 Which means the backlog is not a random sample of private equity. It is structurally the 2018–2022 vintages — bought at the top, financed at zero rates, and still sitting there.
One correction worth making, because it is made constantly: private equity did not fund startups. Venture did that. Private equity bought mature software companies and levered them.
The correct version is worse. In 2021, PE closed $284 billion in technology deals — 25% of total buyout value and 31% of deal count, the largest share of any sector. Software alone was $256 billion of that, 90% of the tech total. One in three buyouts involved a technology company.10
Apply that share to the backlog and you get roughly $1 trillion of PE-held software and technology NAV sitting unexited.
And the debt underneath
The Federal Reserve's May 2026 Financial Stability Report puts U.S. private credit at $1.4 trillion — about 10% of total U.S. nonfinancial corporate debt, and roughly one-third of below-investment-grade corporate debt excluding bank loans.11
The Fed also states plainly that software is now the largest sector in private-credit portfolios, driven by prior private equity activity.11
Congressional Research Service estimated private credit's SaaS exposure at around $500 billion as of December 2025.12 Broader estimates run $600–750 billion, and none of that includes broadly syndicated loans, so the total software LBO debt stack is larger still.
The exposure is not $3.8 trillion. The AI-sensitive envelope is roughly $1 trillion of marked software equity sitting on top of $500–750 billion of debt.
Equity absorbs first. Which is why the equity layer is functionally the whole exposure — right up until it is gone.
Thoma Bravo took Medallia private for $6.4 billion in 2021. It is reportedly expected to lose as much as $5.1 billion.13 A debt load that was trivial at zero rates erased the sponsor's equity outright.
That is the template. Not the outlier.
The calculation that closes the argument
Airtable cleared at 2.7x enterprise value to ARR. Unlevered. Net cash. Growing 20%.
Be precise about what that comp is. Airtable is a horizontal productivity tool selling something agents now do natively — it may be the single most AI-exposed asset in the cohort. Vertical software with regulatory embedding, proprietary data, and captive workflows clears higher. So 2.7x is not the book's average. It is closer to the book's floor.
Run the range instead.
Take the ~$1 trillion of software NAV and add its debt. The claimed enterprise value across that book is somewhere near $1.5–1.9 trillion. At the Airtable floor of 2.7x, supporting that claim requires roughly $550–700 billion of aggregate ARR inside private-equity-owned software portfolios. At a generous blended 4x — pricing every vertical moat at par — it still requires $375–475 billion.
Nobody publishes the actual aggregate. That silence is itself information: it is the one number that would settle the question, and no one on the inside volunteers it.
But the arithmetic bounds the debate. For the marks to hold at the floor multiple, PE-owned software would need combined recurring revenue exceeding all of Salesforce, Microsoft's entire cloud segment, and Oracle put together. At 4x, the requirement drops — and so does the claimed gap, from roughly a trillion dollars to several hundred billion. The generous case does not rescue the marks. It negotiates the size of the write-down.
Reasonable multiples disagree about the size of the gap. No plausible multiple closes it.
Continuation vehicles were built to avoid measuring exactly this. In 2025 the secondary market transacted $233 billion, with $115 billion of it GP-led — up 53% year over year.14 5 Most of that GP-led volume was continuation funds: the same manager on both sides of the trade, setting the price, collecting fees from both vehicles.
That structure is legal. It is also, mechanically, a device for deferring the moment a number gets forced into the open.
The transmission already ran
This is not a forecast. It happened between January and July of this year.
Blue Owl permanently halted redemptions at a retail debt fund in February. Cliffwater capped redemptions at 7% against nearly 14% requested. Blackstone let approximately 8% out of BCRED using balance-sheet capital. BlackRock held its HPS fund to 5% against $1.2 billion in requests.15
In March, JPMorgan marked down software loans held as collateral against its lending to private credit funds, cutting the industry's borrowing capacity.16
Capability release, then software revenue doubt, then bank collateral markdown, then reduced fund leverage, then redemption gates.
Four months. No defaults required.
The Fed spent the May report calling private credit risk "limited and manageable"11 — while reporting in the same document that its own market contacts named AI and private credit among the most-cited risks to U.S. financial stability.17
Both statements are in the same publication. Only one of them is a forecast.
The part that should have been obvious
Look at what Airtable sold.
Build applications without engineers. That was the product. That was the pitch. Non-technical employees, custom apps, no waiting on the engineering queue.
Its entire value proposition was labor compression, and it was priced at 24x forward revenue on the premise that removing developers from the workflow was a scarce and defensible capability.
Agents now do that natively, continuously, and for the price of a subscription to something else.
The category most exposed to AI compression turned out to be the category that was already monetizing compression. Its moat was the scarcity of the thing that stopped being scarce.
And the buyer is not a strategic paying a premium. Bending Spoons is a consolidator that owns AOL and Vimeo,18 whose prior U.S. acquisitions have been followed by significant layoffs.4 That is the terminal state for this cohort: not a growth asset, a cash-flow asset, run with fewer people.
The compression thesis does not require a crash. It only requires that assets financed on the assumption of permanently scarce engineering talent be repriced against a world where that talent is not the constraint.
The steelman, and where it actually lands
The defense of the marks deserves its full weight, so here it is.
Airtable is one company, and arguably the worst-positioned one in the entire cohort. The book is not all horizontal tools; much of it is vertical software wired into regulated workflows, where the revenue is protected by something harder than habit. Blend the multiples honestly and the gap shrinks by half or more.
And the system has shock absorbers. Dedicated secondary capital reached a record $327 billion in 2025.5 U.S. private equity holds $880 billion in dry powder.9 Sponsors have both the incentive and the balance sheet to defend marks through continuation vehicles for years. The largest perpetual BDCs can cover multiple quarters of redemptions at the 5% level from bank lines and cash — the Fed checked.11 Deferral is not only evasion. It is also, so far, a functioning buffer that has absorbed every shock thrown at it.
All of that is true. Notice what it argues about.
It argues about the size of the write-down and the speed of its arrival. It does not argue about the direction. A blended 4x multiple still leaves a gap in the hundreds of billions against roughly a trillion dollars of marked equity. The secondary capital is absorption capacity, not price support — it buys tail-end funds at 73 cents, not at par.5 And "sponsors can wait" is not a rebuttal of this piece. It is this piece. Waiting is what the continuation vehicle is.
The bull case, examined closely, is the base case with better manners: the same repricing, spread over more years, with fees collected along the way.
The honest range of outcomes runs from a grinding multi-year workout absorbed almost entirely by LP equity — the likeliest path — to a gate-driven spiral where forced sales set comps faster than the deferral machinery can meter them. The Q1 redemption wave was a preview of the second path.15 What no plausible path includes is the marks being right.
Nobody is arguing about price
U.S. software private equity deal value ran $16.24 billion through the first five months of 2026. That pace annualizes to roughly a quarter of 2025's record $156 billion.
Platform deals accounted for 41% of software PE deal value — the lowest share in at least a decade, down 30 percentage points year over year. Seven software platform deals have cleared $100 million all year.13
Buyers have not disagreed on valuation.
They have stopped bidding.
That is the condition under which marks become fiction, and it is the condition under which every subsequent transaction becomes a comp against $3.8 trillion of inventory that has to move eventually.
Airtable was the good outcome. It had growth, enterprise penetration, brand, and no debt. Equity holders received something because there was nothing ahead of them in line.
Now run the same multiple compression through a company at five turns of leverage.
The equity is not impaired. It is absent.