The SaaSpocalypse Is a Vibe

Last week Salesforce reported what its CEO called one of the best quarters in the company's history: $11.3B in revenue, up 11%, free cash flow of $15.2B over the trailing year, customer attrition near an all-time low, and $3.9B of annual recurring revenue from Agentforce and Data 360, the AI products that were supposed to be the evidence of its obsolescence.¹

Six months earlier, the market had priced this company, and most of its industry, for something close to death. In February a Jefferies trader coined a word for it, the SaaSpocalypse, and the word stuck because the tape agreed: roughly $285B of software market value erased in one 48-hour stretch, an index of 120 public SaaS companies down 25.7% in a single quarter, JPMorgan marking down software-backed loans, and more than $10B of redemption requests hitting private credit funds whose collateral was, in the end, subscription software.²³

So we have a puzzle worthy of a long essay. The narrative says AI agents are eating software. The income statements say software is having one of its best years ever. Both cannot be fully true, and I have learned that when a narrative and a ledger disagree, you audit the ledger first. So I pulled the numbers myself: quarterly financials, market data, and valuation multiples for 14 public SaaS companies, from the giants to the walking wounded, as of August 27, 2026. This essay is that audit.

Almost everything in the apocalypse turns out to be real except the death. The dying is happening somewhere else on the financial statements than everyone is pointing.

Every panic has a lineage, and this one is unusually traceable. In December 2024, Satya Nadella said on the BG2 podcast that SaaS applications as we know them will collapse in the agent era, because the business logic that applications wrap around databases would migrate to the AI layer.² It was a thoughtful architectural argument from the person who sells more SaaS than anyone alive, and it detonated slowly. Through 2025 the anecdotes stacked up: Klarna announcing it was ripping out Salesforce and Workday to replace them with AI, then quietly walking parts of that back the following year. Startups with terrifying velocity, Harvey against legal research at an $11B valuation, Sierra and Decagon against customer service, Cursor turning every enterprise into a potential builder of its own tools.²

Then, in January and February of this year, the foundation model companies raised about $160B in a few weeks, and the market did what markets do with a story that suddenly has that much capital behind it. It stopped debating and started selling. The Jefferies coinage gave the fear a name, the name gave it a chart, and by the end of March the SEG SaaS index was down a quarter of its value in ninety days.² The hardest-hit names were exactly the ones the thesis predicted: horizontal seat-based platforms like HubSpot, Atlassian, and Workday, where an agent could plausibly do what the seats do.

I want to be fair to this panic, because I think it was a reasonable idea executed with the reasoning depth of a stampede. The bear case has real content, and I will give it its own section. But first, the part almost nobody in February bothered to check: what the businesses were actually reporting.

Here are 14 public SaaS companies, chosen to span the spectrum: the CRM and ERP giants, the data and observability layer, security, commerce, and the seat-based tools the apocalypse thesis says should die first. For each, the year-over-year revenue growth in the most recently reported quarter, against the growth rate in the same quarter a year earlier.

Read that chart slowly, because it is the whole myth in one picture. All 14 companies grew. Eleven of the thirteen with comparable year-ago data grew as fast or faster than they did a year before. Datadog accelerated from 28 to 36%. Snowflake from 26 to 34. MongoDB, CrowdStrike, ServiceNow, all accelerating. These are not the vital signs of an industry being eaten. Two and a half years into the age of agents, the median company in this basket is growing faster than it was when the SaaSpocalypse was coined.

The profitability picture is stronger still. Across the public SaaS universe, median EBITDA margins swung from negative in 2022 to solidly positive, median net income turned positive in 2025 for the first time in years, and 93% of S&P 500 software companies beat profit estimates in the quarter the panic peaked, against 74% for the broader index.² Deutsche Bank's strategists, after canvassing the industry mid-panic, reported that they could not find a single software company expecting a negative revenue effect from AI in 2026.²

And the AI that was supposed to be the disease keeps showing up as revenue. Salesforce's agent products crossed $3.9B in ARR, with 3.2 billion agentic work units consumed in the quarter, up 97% sequentially, and a six-fold increase in MCP calls, which is a detail I will return to.¹ Zendesk projects half a billion dollars of AI recurring revenue by year end.⁴ The most on-the-nose data point of all: the company whose churn the agents were supposed to cause reports customer attrition near an all-time low.¹

Now the other half of the audit, because the market is not stupid and the panic did not fall evenly. Here are the same 14 companies, ranked by how far below their 52-week high they traded at the end of last week.

This chart is the market thinking out loud, and what it is thinking is not "SaaS is dead." It is something far more surgical. At the top: CrowdStrike at an all-time high, Snowflake and MongoDB within a few points of theirs, Datadog close behind. The market has decided that the data, security, and observability layer wins in an agent world, for a reason that any builder could have told you: agents are software that consumes software. Every agent that replaces a human worker still logs, still queries, still authenticates, still gets attacked. The picks and shovels of the agent economy are mostly SaaS companies.

At the bottom: HubSpot and Intuit cut in half, Asana down a third, ServiceNow down 29% from its September peak even while growing 24%. This is where the thesis lives. The market is discounting exactly the businesses whose value proposition is a human sitting in front of a licensed screen, and it is doing so almost without regard to their current growth. HubSpot grew 20% last quarter, the same as a year ago, and lost half its market value anyway.

Which brings us to the multiples, where the apocalypse actually happened.

Sit with those numbers the way I had to. Zoom trades at 8.7 times earnings, the multiple of a regional bank. Adobe, a franchise with 89% gross margins growing 13%, trades at 15.6 times, cheaper than Coca-Cola. Salesforce, after its blowout quarter, trades at 18.7 times trailing earnings, roughly the multiple of the average industrial. These are the multiples of companies the market believes have stopped compounding. Not dying today, not even shrinking, but finished growing in any way that matters to a discounted cash flow that extends past 2030. The apocalypse was never in the revenue line. It is a terminal value story, priced entirely into the multiple.

One company makes the whole phenomenon legible, so let me put five and a half years of it in a single picture. Salesforce revenue and Salesforce stock, both indexed to January 2021.

Since January 2021, Salesforce has slightly more than doubled its trailing revenue, from about $21B to about $44B, while roughly tripling operating cash generation and swinging from thin profitability to $3.5B of quarterly net income. The stock, over the same five and a half years, has returned about 12%. Not 12% a year. Twelve percent, total, before dividends. The price-to-sales multiple did the reconciling: from around 10 times sales to under 5.

There are only two coherent readings of that chart. Either the market of 2021 was wrong about what this business would be worth, or the market of 2026 is wrong about what it will become. My honest answer is: mostly the former, partly both. 2021 priced SaaS as if 25% growth were a law of nature; the actual decade delivered a maturing industry, and the multiple came down to meet reality. That part is not an apocalypse. That is arithmetic. The 2026 part is the interesting one: the market is now applying a disruption discount on top of the maturity discount, to the exact companies reporting record AI revenue. One of those two discounts is being applied twice.

A stock that goes nowhere for five years while the business doubles is not a verdict on the past. It is a bet about the future, made with other people's conviction.

Now the part the myth-busting genre always skips, because I promised to be fair to the bear case, and the bear case is not stupid. Three things in the apocalypse story are true, and the chart of the drawdowns shows the market has noticed all three.

First, the industry really did downshift. Median public SaaS revenue growth fell from 27% in 2021 to about 10% in 2025, and net revenue retention slid from 117 to 108.² That decline predates agents and has more to do with maturity and post-pandemic normalization than AI, but it means the industry met this narrative with less momentum than at any point in its history. My basket looks better than the median because public markets have already sorted survivors from the long tail; the tail exists, it is just private, or small, or both.

Second, the seat is genuinely in trouble. Look again at the growth chart's bottom three: Zoom at 4.9%, DocuSign at 8.7, Asana at 9.5. These are fine, profitable, slowly-compounding businesses, and they are also exactly what the thesis predicts: products whose pricing is a count of humans, in categories where the work itself is being automated. When Bain estimates AI could automate 30 to 50% of activity across enterprise functions, the seat-count exposure is not hypothetical.² The companies escaping the discount are the ones that changed what they meter: Salesforce now sells agentic work units, Snowflake and Datadog sell consumption, CrowdStrike sells protection per workload. The apocalypse is not coming for software. It is coming for the seat as a unit of account.

Third, the buyer has an alternative for the first time. Harvey, Sierra, Decagon, and a hundred smaller versions are real companies with real revenue, and coding agents genuinely lower the cost of building internal tools that once justified a subscription. I run experiments on this every week; the build-versus-buy line has moved. What I do not believe, having also operated the other side, is that it moved as far as the valuations imply. A tool is an afternoon. A system of record with security, compliance, uptime, and ten years of your own data inside it is not, and as one analysis of the panic put it, the screen may be optional now, but the capability is not.⁴ The six-fold growth in MCP calls hitting Salesforce is the tell: agents are not replacing the system of record, they are becoming its heaviest users.¹

The table rewards a slow read. Notice that the price-to-sales column no longer correlates with the growth column the way a 2021 spreadsheet would predict. HubSpot grows at 20 and trades at 3.7 times sales; CrowdStrike grows at 26 and trades at 43. The market is not paying for growth anymore. It is paying for position in the agent stack, and it is doing so with a confidence about which layer wins that I do not think anyone has earned yet.

So is the SaaSpocalypse a myth? Here is my verdict, stated the way I would state it in a board meeting.

As a claim about businesses, yes, it is a myth, and the audit above is the evidence: fourteen for fourteen growing, most accelerating, margins at record highs, attrition at record lows, and the supposed disease showing up as billions in new revenue. The strong version of the thesis, the one where agents dissolve the application layer within a few years, has so far produced exactly zero confirmed instances of a large enterprise actually completing the replacement. Even Klarna, the poster child, ended up somewhere between partial and embarrassed.

As a claim about prices, it was not a myth at all. It was one of the fastest repricings of an entire sector I have watched in my decades around this industry, and it left a body count: a quarter of the sector's value, marked-down loan books, gated credit funds, and a private equity complex sitting on software bought at 6.7 times revenue that now marks at 3.1.² Repricings of that violence change behavior regardless of whether the thesis behind them is right. Boards cut, founders sell early, and the cost of capital for anything seat-shaped goes up. In that sense the SaaSpocalypse is partly self-fulfilling, which is the most honest thing you can say for it.

And as a claim about the future, it is a genuinely open question that the market has chosen to answer with certainty it does not possess, in both directions at once: pricing the data layer as untouchable and the application layer as terminal, when the actual evidence, the MCP calls, the agentic work units, the attrition figures, suggests something less cinematic. The application layer is not dying. It is being renegotiated, from seats to work, from screens to capabilities, from humans clicking to agents calling. Some franchises will not survive the renegotiation. The seat-shaped ones, the ones at the bottom of my growth chart, probably will not, at least not in their current form. But renegotiation is not apocalypse, and the difference between those two words is currently worth a few hundred billion dollars of mispricing, in one direction or the other.

I keep coming back to the meta-lesson, because it is the same one I took from the enterprise AI adoption numbers earlier this year: narratives now move at the speed of a funding round, and income statements still move at the speed of a fiscal quarter. The gap between those two clocks is where fortunes get made and lost, and where, for those of us who build rather than trade, the only defensible strategy is to keep auditing the ledger while everyone else audits the vibes. The SaaSpocalypse was a vibe. The renegotiation is real. Confusing the two was the most expensive category error in software this year.

Company financials are from public quarterly reports through August 27, 2026; prices, market capitalizations, and multiples are from market data as of that date. Nothing here is investment advice; it is an essay about the distance between stories and ledgers.