The SaaSpocalypse Has a Loophole

Public SaaS has lost over $1 trillion in market value since the 2021 peak. The iShares Software ETF is down 21% year-to-date. Median EV/revenue multiples have collapsed from 18-21x to 5-7x. The average public SaaS name is down 32% over the past twelve months, a drawdown worse than COVID.

The consensus narrative is tidy: AI eats software. Seat-based pricing dies. Margins compress. SaaS is over.

And the 2025 scoreboard seems to confirm it -- until you look closer. SaaStr documented the split: Palantir finished the year up 142%. Cloudflare up 80%. MongoDB up 70%. HubSpot dropped 51%. The gap between winners and losers was the widest in SaaS history, and the line separating them had nothing to do with growth rates or rule-of-40 scores or any of the metrics that mattered eighteen months ago.

The line was whether the company had figured out how to make AI the actual product -- not a chatbot in the corner of the dashboard, but the thing the customer is paying for.

Most SaaS companies in 2025 bolted on a GPT wrapper and called it a strategy. The ones that rerated did something harder. They tore apart their own product, rebuilt the core workflow around automation, and in several cases changed the pricing model entirely. They used the crash as cover to do things they never would have attempted when the stock was at $40.

This is playing out right now. And one company, in particular, shows what the full arc looks like.

Blend's Autopilot

Blend Labs went public in June 2021 at a $4 billion market cap. The company, founded in 2012 by Nima Ghamsari, built white-label software powering digital mortgage applications at major U.S. banks. The investor list was a fintech Hall of Fame: Greylock, Founders Fund, Andreessen Horowitz, Lightspeed, Temasek, General Atlantic. The pitch was broad -- a platform that could digitize everything from mortgages to consumer credit products -- and during the ZIRP-era refinancing boom, the numbers were spectacular.

Then interest rates spiked. Mortgage volumes cratered. First Republic, one of Blend's biggest customers, collapsed entirely. The stock fell over 90%. Market cap dropped to around $437 million.

"It probably gave me an inflated sense of how well I was executing," Ghamsari told Fortune. "I had overestimated my operating ability." He said he'd made the company take on too many things during the boom. The sprawl that venture capital rewarded became the liability that public markets punished.

What Ghamsari did next is where the story gets interesting, and where it diverges from the typical post-crash narrative of slow decline and eventual acquisition.

He didn't pivot. He didn't chase a new market. He went backward -- stripped the company down, cut multiple rounds of layoffs, narrowed the focus to being "really, really great at one thing," and forced the business to profitability. By Q4 2025, Blend had posted five consecutive profitable quarters. Revenue was $32.4 million, up 7% year-over-year. The consumer banking suite was growing 21%. Nobody was writing headlines about it.

Then, in March 2026, Blend launched Autopilot.

Autopilot is an AI agent that reads borrower documents, checks them against lender compliance rules, updates the loan file, and kicks off follow-ups with borrowers -- autonomously. Loan reviews that took days now complete in 15 seconds. Humans and existing compliance systems still make the final call. But the manual labor -- the document shuffling, the data entry, the back-and-forth emails -- is gone.

Ghamsari frames this around a specific number: roughly $11,000 in human cost and hundreds of hours per mortgage, which lenders currently eat. Autopilot is a direct attack on that cost. Within a month of launch, 20% of Blend's customers had adopted it.

This is a different company than the one that IPO'd. The 2021 version of Blend sold a nicer interface for filling out forms. The 2026 version sells an autonomous workflow that eliminates entire cost centers. The product changed, the margin profile changed, and eventually the valuation framework should change too -- though the stock, still trading in the low single digits, hasn't caught up yet.

"The hardest thing about the layoffs is you still believe in the business," Ghamsari said. "You just feel like you did the wrong things that led us to the point of the layoffs." He argues that owning those decisions publicly -- rather than spinning them -- is what made it possible to rebuild the culture around the new product.

Whether Autopilot alone can rerate Blend is an open bet. But the shape of the story -- peak, crash, brutal reset, AI rebuild on top of existing distribution and data -- is showing up elsewhere.

The Same Shape, Different Industries

Palantir's version of this arc is the most dramatic, and the one that probably should have tipped everyone off that the pattern was real.

In 2023, Palantir was growing at 13%. Twenty years old, perennially controversial, stuck in a narrative about government contracts and Alex Karp's eccentricities. The stock was dead money. Then the company launched AIP -- its Artificial Intelligence Platform -- and started running intensive "boot camps" where enterprise customers deployed Palantir on real problems in days instead of months. The results were absurd. Revenue growth reaccelerated to 70% year-over-year by Q4 2025. U.S. commercial revenue surged 137%. The Rule of 40 score hit 127 -- a number that would be impressive for a Series B startup, let alone a company old enough to vote.

Palantir didn't build a new product so much as reframe the old one. Ontology, the data integration layer the company had spent two decades building, turned out to be exactly the kind of proprietary infrastructure that AI models need to operate inside a large enterprise. The moat they'd been quietly digging since 2003 suddenly had water in it.

MongoDB's turnaround was quieter but just as real. Growth stalled in 2024. Founding CEO Dev Ittycheria stepped down. Analysts wrote it off. Then something shifted: developers building AI applications needed databases that could handle unstructured data -- embeddings, vectors, document stores -- and MongoDB's architecture was a natural fit. Atlas revenue jumped 30% year-over-year in Q3, the stock climbed 70%+ through 2025, and total revenue crossed $628 million for the quarter. MongoDB didn't pivot to AI. The AI wave pivoted to MongoDB. Sometimes reinvention is just being in the right place when the water rises.

And then there's Sabre, which is attempting something more radical than any of them.

Sabre is a travel technology company most people have never heard of, even though it quietly powers a large share of global airline and hotel bookings. The stock is down over 80% from pre-pandemic levels. COVID gutted the travel industry, and Sabre's legacy GDS infrastructure -- the backbone of its business -- looked increasingly like a relic.

In March 2026, at the ITB Berlin trade show, the company unveiled what its leadership called a "once-in-a-generation company rebuild": a unified, AI-native cloud platform replacing decades-old architecture. Agentic workflows for booking. Autonomous retailing. Enterprise AI governance baked in from scratch. They're not adding AI to the old system. They're throwing the old system away and building a new one that assumes AI agents, not human travel agents, will be the primary users.

Five years in the making. Whether it works is genuinely unclear. But the scale of ambition tells you something about how far this idea has traveled. When a 60-year-old travel infrastructure company decides to rebuild itself from zero around AI, the thesis isn't niche anymore.

The Buyout Version

Here's what makes this bigger than a handful of turnaround stories.

If the playbook works when a founder does it to his own company under duress -- Ghamsari at Blend, staring at a 90% stock decline and deciding to rebuild rather than sell -- then it works even better when someone with fresh capital does it deliberately to a company that's too stuck, too bureaucratic, or too demoralized to do it to itself.

Private equity figured this out about eighteen months ago. And they haven't been subtle about it.

Thoma Bravo, the firm that built a $130 billion empire buying enterprise software and squeezing out inefficiency, has started adding AI directly to the operating playbook. In late 2025, they acquired Verint for roughly $2 billion, merged it with Calabrio, and positioned the combined entity as an AI-powered customer experience automation platform. Separately, Thoma Bravo took Dayforce private for $12.3 billion -- a workforce management company with an obvious AI automation layer waiting to be built on top of its existing payroll and HR workflows. The old Thoma Bravo thesis was: buy mature software, cut costs, expand margins, exit. The new thesis appends one step: insert an AI layer that structurally lowers cost-to-serve. It's the same leveraged buyout math, but with a new variable.

Jared Kushner's Affinity Partners, now managing $6.2 billion backed heavily by Saudi PIF, is running a splashier version. The firm co-led the $55 billion take-private of Electronic Arts alongside Silver Lake and PIF last September -- the largest LBO in history. And they co-founded Brain Co., a $30 million AI startup built in partnership with OpenAI, already selling to clients like Sotheby's and Warburg Pincus. Kushner's play is less traditional PE turnaround and more sovereign-capital-meets-AI-overlay: buy large assets with global capital, retrofit them with AI capabilities that wouldn't have been possible three years ago.

The PE angle matters not because it's a different thesis. It matters because it's the same thesis with institutional-scale capital behind it. When Thoma Bravo is spending $14 billion on this idea, it's not a hot take on a blog. It's a capital cycle.

The Math, and the Doubt

The traditional SaaS value creation playbook was simple: grow revenue fast, expand the multiple, exit. That playbook is broken at 5-7x multiples and 8% growth rates.

The AI reinvention playbook attacks a different line item. It goes after cost structure, not the top line.

Take Blend's math. The mortgage industry spends roughly $11,000 in human labor per loan origination. Document review, verification, compliance checks, borrower follow-up. Hundreds of hours. If Autopilot can automate 30-40% of that workflow, the savings run $3,000-$4,000 per loan. Blend doesn't need the mortgage market to recover to benefit. It needs to capture a piece of the cost it's eliminating.

Palantir does the same thing to consulting labor. Enterprises pay for the platform instead of paying Accenture for the bodies. Thoma Bravo's Verint deal bets that AI can replace human call center agents, converting a services-heavy P&L into a software-margin P&L. Sabre is rebuilding so that AI agents replace the human workflows in airline retailing and hotel distribution.

It's margin engineering, not revenue growth. And in a market that rewards profitability and free cash flow, margin expansion is the trade that gets repriced.

Now, the doubt.

Whether these companies actually capture the savings they create is not settled. Blend can cut $4,000 from the cost of originating a mortgage, but that doesn't mean Blend gets paid $4,000 more. Lenders might pocket the savings and squeeze the vendor. Palantir's margins are extraordinary, but its customer concentration is still high and government contracts are fickle. Sabre's rebuild is a five-year bet on architecture that hasn't been proven at scale. And the PE playbook -- bolt AI onto acquired software, reprice the multiple -- only works if the AI actually changes the product and not just the investor deck.

There's a real risk that half of what's being marketed as "AI transformation" is just rebranding with better pitch decks. The gap between companies doing the Blend thing -- where the AI is the product, running autonomously, replacing real labor hours -- and companies stapling a chatbot to an existing SaaS product and calling it AI-native, is enormous. The market doesn't always distinguish between the two. At least not immediately.

Who the Loophole Is For

If you accept the pattern -- public SaaS company, stock down 60-90%, strong distribution, embedded in complex workflows, high human labor costs in the customer base -- the candidates start to multiply.

nCino (NCNO), the bank operating system, stock down 46% over the past year, sitting on the same mortgage-adjacent plumbing as Blend but broader across commercial lending. Domo (DOMO), the BI platform that cratered, sitting on enterprise data that's one AI layer away from being an analytics agent. Paylocity and Paycom, compressed HR and payroll companies running repetitive, rule-based processes that are textbook automation targets. The entire legal software vertical -- contract review, discovery, compliance -- beaten down and perfectly suited for agentic automation.

These are all companies with real revenue, real customers, real distribution, and cost structures that AI can restructure. The market is pricing them as declining SaaS. Someone -- a founder who gets it, a PE firm with operating capital, a strategic acquirer -- is going to reprice them as AI-automated workflow businesses. The spread between those two valuations is where the money is.

The SaaSpocalypse is real. A trillion dollars in SaaS value has been destroyed and the old playbook isn't coming back.

But the wreckage contains something useful: mature software businesses with distribution, proprietary data, and customer relationships that took a decade to build. The companies and investors who see that -- who recognize that the cheapest way to build an AI business might be to buy a broken SaaS business and rebuild it from the inside -- are playing a different game than the people writing eulogies.

Blend's Ghamsari said it clearly: he overestimated his own abilities during the boom. The crash forced him back to first principles. What he found there was an AI agent that does in 15 seconds what used to take his customers days.

The trillion dollars isn't gone. It's being repriced. And some of it is about to come back wearing a very different operating structure.