$26T

The $26T of the Software Innovators Leaderboard is not a single bucket. It splits into four ownership cohorts, Public, PE-owned, VC-private, and Acquired. The Public cohort is the only one working as designed, repricing in plain sight on quarterly earnings calls. The other three are each broken in a different way: PE-owned is breaking on the Medallia mechanic, VC-private is trapped behind a velvet rope, and Acquired is being misread as validated value when most of it is failed independence. For an investor, owner, or operator of a software company, the cohort an asset sits in is now doing more of the work than the category it competes in. This piece walks each.

The 218 companies in the public cohort carry $18.0 trillion of value, sixty-nine percent of the leaderboard, and the only cohort an individual investor can actually touch. It is also the cohort the AI cohort is repricing first, because public companies are obliged to mark their position every quarter and obliged to act on it where the math no longer works.

Every SaaS-era public name is now sitting in front of the same productivity comparison, against the same AI-native cohort, with the same set of obligations to its shareholders. The adjustments will arrive one at a time, on quarterly earnings calls, over the next two years, with the named Wave IV substitutes (Cursor, Hex, Glean, Clay, and the three frontier labs behind them) already winning the net-new line items where they compete. The public cohort is the cohort where the repricing is most legible, because it is the cohort where it has to be disclosed.

The mechanics matter, because the same mechanics apply to roughly a dozen other deals on the same shelf.

Thoma Bravo had taken Medallia private in 2021 for $6.4 billion. By April of this year, the equity was worth roughly zero. Blackstone, KKR, Apollo, and Antares — the private credit lenders sitting senior to the equity — now own the company. Thoma Bravo and its co-investors wrote off roughly $5.1 billion. The trigger was not a missed quarter or a sudden customer loss. It was the expiration, at year-end 2025, of a Payment-in-Kind relief arrangement that had let Medallia defer its cash interest by piling it onto principal. Once that mechanism switched off, the cash interest bill arrived in full, against revenue and EBITDA that had not grown the way the 2021 underwriting assumed. The capital structure failed. The company kept operating; the equity went to zero.

Medallia was the first big one, and it will not be the last. Across the 456 enterprise software companies that sit in private-equity hands in our register, combined enterprise value $1.1 trillion, there is somewhere in the neighborhood of $46.9 billion in distressed software loans currently in the private credit market, according to Moody's. A meaningful share of that capital was deployed during the 2021 vintage, at peak multiples, with deal structures that financed up to half of the purchase price with debt at rates that only worked when interest rates were near zero.

Those rates are no longer near zero. The expansion math underwriting the debt has not held. And the AI cohort underneath the SaaS layer is now compressing the very expansion math the LBO models needed to clear.

The named PE-held software companies the trade press has flagged in 2026 carry debt structures that depend on growth or refinancing assumptions the current rate and category environment is not delivering. Numbers below come from the Airframe register; debt and lender details from S&P Global Market Intelligence, SaaStr's analysis, and Octus.

That is roughly $46 billion of equity-plus-debt exposure across seven named deals, and it does not include the dozens of mid-sized deals in the same vintage carrying the same structural issues at smaller scales. SaaStr counts at least twelve more deals with combined debt exposure north of $50 billion at meaningful risk over the next two years.

A leveraged buyout of an enterprise software company works on three assumptions: revenue grows modestly, net retention stays above 110%, and the debt can be refinanced when it comes due. All three are bending at once. Revenue growth across the named PE-held names has compressed from the 15–25% range at the time of the take-private to mid-single-digits or, in some cases, flat. Net retention at scale is moving below 100% in categories where it has not been below 100% in a decade. Refinancing the same debt at 2026 rates roughly doubles the interest cost, which is exactly the math that broke Medallia when its PIK relief expired. The companies most exposed are the ones whose debt comes due in 2026, 2027, and 2028. We are now inside that window.

The PE cohort is the one the trade press has narrated for two years. It is the wrong $1 trillion. The larger pool of stuck capital sits in the next cohort over, on the other side of an access regime the previous four waves of software did not have.

Microsoft went public in 1986 at a valuation of roughly $500 million. A schoolteacher with a brokerage account could buy a share that afternoon. Twenty years later, that share was worth more than fifty times what she paid for it. Salesforce was founded in 1999 and listed in 2004 at $1.1 billion. Same story. So is the entire SaaS cohort that followed: Workday, ServiceNow, Shopify, Atlassian, HubSpot, Okta, Twilio, Veeva, Dropbox, Box, Zendesk. Twelve companies, public for most of their growth, now collectively worth roughly $511 billion. Every dollar of that compounding was available to anyone with a discount-brokerage login.

That pattern is over.

The three frontier AI labs at the front of the current cohort — OpenAI, Anthropic, and xAI (now part of the merged X/SpaceX entity) — are now jointly worth roughly $2.1 trillion, and an individual investor with a 401(k) and a Fidelity account cannot buy a share of any of them. Neither can her brother, who's been allocating to public software ETFs for fifteen years. Neither can her financial advisor. Neither, in most cases, can the pension fund that manages her teacher's-union retirement assets, because access to the small private rounds these companies do run is rationed to a handful of repeat venture investors and sovereign wealth funds writing ten-billion-dollar checks. The largest data platform most of the AI cohort runs on is private. The payments platform that moves a trillion dollars a year is private. The companies setting the productivity ceiling of the next decade of software are, in 2026, almost entirely unavailable to the people whose savings powered the last one.

The cost lands on the LPs first. Smaller institutional LPs — university endowments outside the top tier, regional foundations, public-sector pension funds in mid-sized states — are now selling their LP stakes in venture funds on secondary markets at discounts of twenty to forty percent against the fund's stated NAV. They are taking real losses on paper-good positions because the cash is not coming and the bills are.

The $5 trillion locked here is roughly five times the PE cohort that gets the headlines. The trade press has spent two years narrating the $1.1 trillion PE-held cohort as the trapped-capital story of the cycle. It is the wrong $1 trillion. The number that matters is on a different shelf entirely, and it is sitting behind an access regime the previous four waves of software did not have.

A meaningful fraction of the Software Innovators Leaderboard — companies that crossed $500M and were subsequently absorbed — represents failed independence, not validated value. Many were acquired below their peak. Many were folded into larger platforms and effectively disappeared as products. Many were bought to neutralize a competitor rather than to extend a strategy.

The acquired cohort and the public cohort tell different stories. The public cohort is a list of companies that built durable independence. The acquired cohort is a list of companies that built enough leverage to be bought — a meaningfully different and often less impressive achievement. We separate them because the lesson of the dataset depends on the distinction.

This leads to an important fact: each technology era is layered and additive. Each era builds on top of the previous era but the vendor still got replaced. Software has moved through eight phases of disruptive technological change. The instinct is to read these as a sequence of replacements. The reality is that each wave builds on the previous technology era like layered substrates. Mainframe software still runs most of the world's banks today. What gets replaced reliably is the vendor on the line item. The acquired cohort tells that story. In the AI era, no leader on this list is safe.