$5.2T

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 $1.9 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.

Across the 2,406 enterprise software companies in our complete vendor register, $5.2 trillion of enterprise value is trapped in the venture-capital-private stack: 1,109 companies whose valuations are set only at the moments someone chooses to lead a round, and almost none of which offer meaningful secondary liquidity to anyone — employees, early investors, or the limited partners who funded the firms that funded them.

This is not how prior cycles worked, and it is creating a distribution problem the press has not yet caught up with.

It is worth taking the counter-argument seriously. Accreditation rules exist for plausible reasons; retail investors buying OpenAI at a $500 billion mark is not obviously good for retail, particularly if the mark turns out to be wrong by a factor of two. The companies in question are choosing to stay private partly because public-market quarterly pressure would actively destroy the R&D cycles their progress depends on, and that is a legitimate operating consideration. The 1986 disclosure regime was not designed for businesses where the cost of being judged on a ninety-day cycle is genuinely existential. The defenders of the current frame have a case.

The case is incomplete.

For two years, the anxious conversation in software finance has been about the $1.14 trillion of value inside private-equity-held software portfolios — the Veritas, Thoma Bravo, and Vista roll-ups, together with the private credit exposure underwriting the leverage on top of them. That conversation is real, and Airframe covers it in detail in our companion piece on the PE Software Reckoning.

But it is not the largest pool of stuck capital. The number that matters is nearly five times larger.

Of the 2,406 enterprise software companies on our complete register, 387 were founded in the last seven years. Almost all of them live in the VC-private column. The top ten of those names hold roughly $2.64 trillion — roughly half of the trapped stack, concentrated in fewer than a dozen companies.

The pattern for fifty years was: build the company in private, cross some threshold in revenue and product-market fit, file an S-1, and let the public price-discover the rest. The median time from founding to IPO for the SaaS-era pure-plays that eventually listed was about eight years. For the AI-native cohort, the analogous step has not happened for any of the three leaders. OpenAI is nine years old. The cohort behind it is younger and showing no sign of moving toward public listings either.

These companies are choosing to stay private. They have the means to make that choice stick. The cost of the choice falls on someone else.

The first place the cost lands is on the limited partners who funded the venture firms that funded these companies, and the cost there is severe.

A traditional venture portfolio was supposed to return capital to LPs through exits — companies going public, or being acquired at a known price. The LPs who allocated to the 2018, 2019, and 2020 venture vintages were underwriting against an assumption that meaningful chunks of those funds would be liquid by 2026. They are not. The marquee names inside those funds — the ones that would have justified the entire vintage — are exactly the companies that have decided not to IPO. The marks on paper are extraordinary. The cash distributions are roughly nothing.

For institutional LPs at scale, this is uncomfortable but workable. They have other liquidity, other vintages, and the staff to negotiate secondaries when they need them. For smaller institutional LPs — university endowments outside the top tier, regional foundations, public-sector pension funds in mid-sized states — it is closer to a crisis. They have allocation commitments to meet, and the distributions that were supposed to fund those commitments are not arriving. The result is that some of them 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 second place the cost lands is on the people who actually built these companies. Almost none of the 1,109 VC-private names in the register offer meaningful, predictable secondary liquidity to their employees. A decade of option holders at the largest of them are sitting on paper gains they can monetize only through company-sponsored tenders, at company-controlled prices, within company-controlled windows, and often not at all. A software engineer at Microsoft in 1996 who had been there for five years could sell vested shares on any trading day at a publicly observable price. A software engineer at one of the top-three AI labs in 2026, also with five years vested, is structurally less liquid. That is a less competitive form of compensation than the SaaS era offered, and the labor market is starting to price it in.

The third place the cost lands is on the public. This is the part the financial press has been slowest to identify, and it is the largest of the three.

The previous four waves of enterprise software — mainframe, client-server, on-prem, SaaS — produced their winners on the public tape. Microsoft, Oracle, SAP, Salesforce, Workday, ServiceNow, Adobe, Shopify, Atlassian, Snowflake, Datadog, MongoDB. By the time these companies were household names, they were already public. The investor who put $10,000 into Microsoft on the day of its 1986 IPO and held it has roughly $55 million today, after nine stock splits and forty years of compounding. The investor who put $10,000 into Salesforce on its first day of trading and held it has roughly $260,000. The investor who put $10,000 into Snowflake on its first day of trading and held it has roughly $13,000 — disappointing by the prior standard, but participation was at least available.

For the AI-native cohort, the equivalent investor has nothing. There is no equivalent investor.

This is not a small number. The twelve best-known SaaS pure-plays produced $511 billion of public-market value over twenty years. The three AI labs at the front of the current cohort are already worth roughly $1.9 trillion, in less than a decade, with revenue trajectories suggesting both numbers will grow substantially from here. The market has voted on which transition it thinks is larger. The vote is encoded in the marks. The marks are visible only to a small group of accredited and institutional investors who were already inside the rooms where the rounds got priced.

If you are a working professional in your forties, you have probably spent the last decade buying public software through ETFs, watching the SaaS index quietly become twenty percent of your large-cap exposure, and assuming that when the next wave came you would participate in it the same way. The next wave came. You did not participate in it. The cohort that is now repricing the public software you do own is, by a structural choice the cohort has made, not available to you.

That is the distribution story. It is the largest single divergence between the public and private software markets in fifty years, and it is being narrated as a valuation story rather than an access story. The valuation will eventually resolve, in one direction or the other. The access question is the one that will define this cycle for the next decade of retail investors. If the AI cohort comes onto the public tape via a wave of S-1s in 2027 and 2028 — which is what every prior cycle has produced, eventually — most individual investors will buy in at valuations the institutional and accredited cohort has already harvested most of the growth from. If the cohort instead settles into permanent private-market status, individual investors will sit out the wave entirely.

The public tape is not just a trading venue. It is a disclosure regime. A quarterly 10-Q on a public software company tells every other operator — and every retail investor with a brokerage account — what the unit economics look like at scale: revenue growth, net retention, customer concentration, gross margin, free cash flow. None of that exists for the private leaders of the AI cohort. The most valuable tools in the enterprise stack are, in 2026, the least observable ones.

In the absence of the disclosure regime, the next-best version of observability is the structured deployment record itself: what is actually being installed, by whom, at what scale, with what outcome. That is what Airframe builds. We track 17,000+ AI tools across 143,000+ organizations and 114,000+ scored deployments. We see the adoption footprint of every AI-native name in the register against the public incumbent it is repricing. None of that data tells you what OpenAI's net retention is. It does tell you which of the public software names in your portfolio is on the receiving end of the net-new-project shift. That is a working substitute, and right now it is the closest substitute the market has.

The piece of infrastructure most conspicuously missing from the current cycle is not capital. The AI labs have demonstrated they can raise as much capital as they need at the valuations they want. What is missing is a venue — a public-market path that turns the trapped $5.2 trillion into something individual investors can participate in. Every prior wave in our register, from mainframe to client-server to SaaS to cloud, produced its public-market cohort within roughly three to five years of the new category becoming the default answer in enterprise procurement. The current cohort passed that threshold two years ago. The venue has not followed.