Is the "SaaSpocalypse" real?

Why did people suddenly start calling this the “SaaSpocalypse”?

The “SaaSpocalypse” took off because AI started threatening the way SaaS companies make money, rather than simply giving them another feature to sell.

The panic became visible in February, when Anthropic expanded Claude Cowork with tools for legal work, sales, marketing, finance and data analysis. Investors quickly connected the dots. If Claude could do work across several applications, companies might need fewer specialized tools, fewer paid seats, or fewer software purchases altogether.

The market reaction was huge. Reuters calculated that around $1 trillion disappeared from U.S. software and data-services stocks between January 28 and February 5. The S&P 500 Software & Services index fell for seven straight sessions. Thomson Reuters dropped 18% in one day after investors saw how Claude could handle parts of legal and professional work that had previously required specialized products.

The fear then spread well beyond the applications Anthropic had directly targeted. Software investors started asking whether AI agents could bypass dashboards, reduce employee seat counts and let companies build simple tools internally.

That last question has become much harder to dismiss lately. McKinsey's newest global AI survey found that 32% of organizations using AI had already decided against buying at least one software product or feature because they could build the functionality themselves with coding agents.

So the SaaSpocalypse began as a stock-market panic, but the underlying worry has since shown up in actual software-buying behavior.

Was the February SaaS crash mostly an investor overreaction?

Yes. The February SaaS crash priced in disruption far faster than the actual software businesses were deteriorating.

A trillion dollars of market value disappeared within days even though software customers had obviously not cancelled a trillion dollars of contracts during the same week. Investors were changing what they thought future SaaS revenue deserved to be worth.

The latest market action makes the overshoot easier to see. In late August, Salesforce jumped 22.6% after reporting stronger bookings, historically low attrition and continued double-digit revenue growth. ServiceNow, CrowdStrike, Adobe and other software names rallied with it. Axios reported that the State Street Software and Services ETF, which tracks roughly 130 software and IT companies, rose 5.2% in one session and reached a new all-time high.

That rebound does not erase what happened in February. Software stocks fell again during parts of the following months as the same AI fears resurfaced. What has changed is the market's confidence in the speed of the threat.

Investors initially behaved as if Claude, coding agents and custom software could wipe out established SaaS products almost immediately. The latest earnings season makes that timeline look much too aggressive.

The more interesting question today is whether the underlying SaaS economics are quietly getting worse even while the strongest companies keep growing.

What would have to happen for the SaaSpocalypse to be genuinely real?

A genuine SaaSpocalypse would show up in customer spending and SaaS revenue, not mainly in falling stock prices.

We would expect companies to start cutting their software budgets, SaaS revenue to stagnate across a broad group of vendors, existing customers to spend less at renewal, paid seats to shrink as AI reduces headcount, and internally built software to replace purchased products often enough to become visible in industry-wide numbers.

Some of that is already happening.

Benchmarkit's 2026 study of 342 B2B SaaS and AI-native companies found median gross revenue retention falling from 88% to 84%. Vertice found usage-based pricing overtaking per-user pricing for the first time. McKinsey found nearly one-third of surveyed organizations skipping at least one software purchase because coding agents made an internal build possible.

Gartner has gone further. Its research estimates that as much as $234 billion of enterprise application spending could become exposed to what it calls “agentic arbitrage” by 2030, roughly 20% of SaaS spending in that market. Gartner is describing a world where agents complete work across several systems and weaken the relationship between employee count and software revenue.

Those are serious changes.

But a full SaaSpocalypse needs one more ingredient: broad demand destruction. As of now, that part is missing.

Are SaaS companies actually shrinking today?

No. SaaS companies are still growing quickly enough that calling the current market a broad business collapse is very difficult to defend.

We checked the latest reported revenue growth from ten large software companies whose businesses cover CRM, security, collaboration, HR, data, observability and workflow software. Every company in the group was still growing revenue year over year.

The median growth rate was 23%.

Snowflake's latest results are especially useful because they arrived only very recently. Revenue grew 35% to $1.55 billion, product revenue grew 37%, remaining performance obligations reached $9 billion and net revenue retention was still 126%.

Datadog grew 36%. Atlassian grew 28%. CrowdStrike grew 26%. ServiceNow grew 24%. monday.com grew 22%.

Even slower names such as Salesforce, Workday and Okta were still expanding at double-digit rates.

Private SaaS is softer than it used to be, but the same basic picture holds. SaaS Capital's 2026 survey of more than 1,000 private B2B SaaS companies found median growth of 22%. Only 7.3% reported flat or negative growth. For comparison, that figure reached 13% in the much weaker environment measured in 2020.

Software growth has clearly slowed from the easy-money years, but calling it a collapse goes too far.

SaaS company	Latest quarterly revenue	YoY growth
Datadog	$1.12B	36%
Snowflake	$1.55B	35%
Atlassian	~$1.8B	28%
CrowdStrike	$1.47B	26%
ServiceNow	$3.99B	24%
monday.com	$365M	22%
HubSpot	$912M	20%
Workday	$2.65B	12.8%
Salesforce	$11.3B	11%
Okta	$805M	11%

Are companies actually spending less on SaaS now?

No broad SaaS spending retreat is visible yet. Companies are still spending heavily on software, while AI is changing where that money goes.

Zylo's latest SaaS Management Index analyzed more than 40 million licenses and $75 billion of software and cloud spending. Median SaaS spend remained $9,455 per employee.

The striking part sits inside that total. Spending on AI-native applications jumped 108% year over year. At companies with more than 10,000 employees, it rose 393%.

Vertice sees the same pressure from another dataset. Its latest figures put average SaaS spend per employee at $9,324, slightly higher than the previous quarter. Vertice also recorded unusually strong SaaS price inflation earlier this year, which means companies are paying more for software even while procurement teams try to remove waste.

There is plenty of waste to attack. Zylo estimates that 36% of SaaS licenses are unused.

That combination tells us more than the headline spending number alone. Companies have good reasons to cancel marginal applications, remove unused seats and push back on renewals. At the same time, they are pouring money into ChatGPT, Claude, coding agents and other AI-native tools.

The software budget is being fought over much more aggressively. So far, it is still a very large budget.

Is SaaS customer retention getting worse?

Yes. SaaS retention is weakening, and this is one of the clearest places where the underlying pressure is already measurable.

Benchmarkit's latest dataset shows median gross revenue retention dropping from 88% to 84%. Even companies in the 75th percentile weakened, from 95% to 91%.

An 84% gross retention rate means a typical company in that dataset is losing 16% of its existing recurring revenue base through churn or contraction before counting new sales and expansions.

That becomes much harder to hide once new-logo growth slows.

SaaS Capital found the same link from another angle. Companies with net revenue retention between 100% and 110% grew around five percentage points faster than companies in the 90% to 100% range. The businesses with the strongest retention grew far faster again.

This is where the current SaaS market looks noticeably harsher than the aggregate revenue numbers suggest. Customers still buy software, but mediocre products have less protection. They can cut seats, negotiate harder, consolidate tools or build a missing workflow themselves.

The SaaS companies that relied on automatic expansion inside growing customer organizations are feeling that change first.

Is per-seat SaaS pricing starting to break?

Yes. Per-seat SaaS pricing is already losing ground because AI weakens the old connection between the number of employees and the amount of software a company consumes.

Vertice processes more than $75 billion in software spending, so its pricing data gives us a useful view of that shift. In Q1, per-user pricing still accounted for 35% of the pricing models it tracked, slightly ahead of usage pricing at 34%.

One quarter later, usage pricing had risen to 36.5%. Per-user pricing had fallen to 32.4%.

That is a surprisingly fast change for something as fundamental as the way software gets sold.

Benchmarkit's retention numbers explain why vendors are interested. Companies using usage-based pricing recorded median net revenue retention of 108%. Seat-priced companies came in at 98%.

At 108%, an existing customer cohort expands before a vendor signs anybody new. At 98%, it slowly shrinks.

AI makes the distinction more important. A company may eventually employ fewer support agents, developers or salespeople while running millions more automated software actions. Vendors that charge for transactions, tokens, workflows, compute or agent activity can monetize that growth. A vendor billing only for human logins has fewer options.

We are already seeing incumbents adapt. Salesforce now talks explicitly about Agentic Work Units. Atlassian is building AI consumption into its commercial model. Snowflake has always been much closer to usage pricing than traditional seat-based SaaS.

The seat model will survive in plenty of products, but its position as the default way to monetize enterprise software is clearly weakening.

Pricing model	Share in Q1 2026	Share in Q2 2026
Usage-based	34.0%	36.5%
Per-user	35.0%	32.4%
Hybrid	31.0%	31.1%

Are companies really building software instead of buying SaaS?

Yes. Companies are already skipping real software purchases because AI coding tools have made internal development much cheaper.

McKinsey's newest State of AI survey is the strongest evidence we have because it asks directly about purchasing behavior. Among organizations regularly using AI, 32% said they had decided against buying at least one additional software product or feature because they could build the functionality internally with agentic coding tools.

The percentage reached 41% in technology and 39% in healthcare.

Retool found an even larger shift among its much more builder-heavy audience. Its survey of 817 people found that 35% had already replaced functionality from at least one SaaS product with custom software, while 78% expected to build more internal tools this year.

Retool obviously sells software for building internal applications, so its sample should be treated accordingly. The McKinsey result makes the pattern much harder to wave away.

The interesting detail in Retool's survey is what people are replacing. Workflow automation was the most exposed category at 35%, followed by internal admin tools at 33%, BI tools at 29%, CRM and form-builder functionality at 25%, project management at 23% and customer support at 21%.

That hierarchy makes sense. AI has dramatically reduced the effort needed to build dashboards, approval flows, CRUD applications, internal databases and narrow workflow tools.

Production software still requires testing, permissions, maintenance and security. Retool's own numbers make that clear: only 31% of builders who had shipped software were prompting their way to complete apps. Most were using AI to write pieces of code that humans then integrated into larger systems.

The change is practical. A company evaluating a $30,000-a-year internal workflow product now has a credible third option alongside “buy it” and “live without it.”

It can build the useful part itself.

Can AI agents really make SaaS interfaces less important?

Yes. AI agents can increasingly sit between the user and the SaaS product, which makes the dashboard itself less valuable than it used to be.

Gartner calls the process “agentic arbitrage.” Instead of an employee opening five applications and moving information between them, an agent can retrieve data from those applications, perform the work and return the result.

The underlying software can still be crucial. The human may simply stop visiting it.

Salesforce's recent Anthropic partnership shows what this looks like in practice. Claude can work with Salesforce data, business logic and actions through dozens of prebuilt skills. A salesperson can ask Claude to inspect accounts, reason over customer information or update work without treating Salesforce's traditional interface as the starting point.

That arrangement creates a strange situation for incumbent SaaS companies. Their underlying systems can become even more useful to machines while their visible product becomes less central to the user.

Atlassian is seeing another version of the same behavior. Rovo-assisted actions grew more than 50% quarter over quarter in its latest reported quarter, and more than 80% of Fortune 500 companies now use Rovo. Atlassian says customers using Rovo complete 20% more Jira work items and create or edit 25% more Confluence pages.

Those customers are using more Atlassian infrastructure because AI is involved.

The competitive question is shifting from “will people keep opening our app?” toward “will agents still need our data, permissions and actions?”

For many SaaS companies, that second question is tougher.

Which SaaS products are most exposed to the SaaSpocalypse?

Thin workflow tools, generic point solutions and products with weak data moats look most exposed to the SaaSpocalypse today.

We can get surprisingly specific here by combining the evidence.

Retool's replacement survey puts workflow automation, admin tools and BI dashboards near the top of what companies are rebuilding. Zylo finds 36% of SaaS licenses already going unused, so procurement teams have plenty of obvious targets. McKinsey now shows that internal development is actively influencing software purchases. Gartner expects agents to bypass UX-heavy applications across multiple systems.

The dangerous combination is a product where most of the customer value comes from a fairly simple interface wrapped around logic the customer can reproduce.

A dashboard that queries existing company data is easier to challenge than the database holding the authoritative data.

An approval tool is easier to recreate than the identity system determining who is allowed to approve something.

A lightweight support layer is easier to replace than the underlying customer history, permissions and integrations feeding it.

We can see the other side of this split in current earnings. Datadog grew 36%, Snowflake 35%, CrowdStrike 26% and ServiceNow 24%. These companies sit close to infrastructure, security, data or deeply embedded enterprise processes.

Workday offers another useful clue. More than 5,500 customers are now using at least one of its internally developed agents, and AI accounted for more than 25% of new ACV in its latest quarter. Customers appear willing to let agents operate on top of the HR and finance system they already trust.

The strongest protection currently comes from owning something an agent needs: authoritative data, permissions, security, transaction rails, accumulated workflow context or a deeply embedded system of record.

A pretty dashboard is becoming much less reassuring.

Are big SaaS companies actually making money from AI?

Yes. AI has started generating enough revenue inside several major SaaS companies to move their financial results.

Salesforce is the clearest example right now. Agentforce ARR has passed $1.5 billion and grew more than 240% year over year. Agentforce plus Data 360 reached nearly $3.9 billion of ARR. Salesforce also reported 3.2 billion Agentic Work Units during its latest quarter, almost twice the previous quarter.

ServiceNow says its AI products have crossed $1 billion in annual contract value while total revenue is still growing 24%.

monday.com reported that ARR from AI products doubled sequentially and represented 17% of net new ARR in its latest quarter.

Workday says AI drove more than a quarter of new ACV.

These figures need a little care because every company defines its AI revenue differently. Salesforce recently broadened what it includes inside Agentforce ARR, for example. A dollar of “AI ARR” across two vendors may describe very different products.

The scale is still meaningful. We have moved past the period where every SaaS company could launch a chatbot, call itself an AI company and leave investors guessing about whether anybody paid for it.

At several incumbents, customers clearly are paying.

That gives the strongest SaaS vendors a route out of the seat-pricing problem. They can lose some traditional seat growth while charging for agents, data usage, actions and premium AI functionality.

Are AI costs wrecking SaaS profit margins?

No, at least so far. AI costs have become a real expense for SaaS vendors, but broad software margins are holding up much better than the bearish thesis would suggest.

Benchmarkit's latest research found median software gross margin still above 80%, roughly stable across four years. Its data shows no industry-wide gross-margin collapse from AI infrastructure costs yet.

The public companies provide useful checks.

Salesforce's latest non-GAAP operating margin reached 34.1%. Workday reached 31.1%, up from 29% a year earlier. HubSpot moved from a GAAP operating loss to a 4.8% GAAP operating margin while revenue grew 20%. monday.com reported record non-GAAP operating income in its latest quarter.

AI does create new costs. Tokens, inference and compute can behave very differently from the near-zero marginal cost investors loved in traditional SaaS.

McKinsey's new survey found that operating costs have already constrained some companies' use of software coding agents. Vertice also finds that consumption-based software bills can swing sharply from month to month.

For now, those costs are being absorbed alongside better employee productivity, restructuring and new forms of monetization.

We would change our view quickly if gross margins started falling broadly while AI usage accelerated. The current numbers point in the opposite direction.

Have SaaS valuations permanently changed?

Yes. SaaS valuations have undergone a real reset, and this part of the SaaSpocalypse is much more convincing than the idea that software revenue itself is disappearing.

The SaaS Capital public index reached a median ARR multiple of 16.9x at the 2021 peak. By the end of last year, the median had already fallen to 5.6x.

Then came the AI repricing.

The median dropped to 3.7x during the first quarter and reached 3.2x in June, its lowest reading since 2011. It recovered to 3.8x in July, and larger cloud stocks have rallied sharply since, but the old valuation regime remains far away.

Growth now makes an enormous difference.

An analysis of SaaS Capital's June constituents found companies growing between 20% and 30% trading at a median 5.5x revenue multiple. Companies growing between 10% and 20% were around 3.1x. Companies below 10% growth were down around 1.9x.

Recurring revenue by itself used to earn investors' respect. Today they want growth, retention, margins and some reason to believe AI will strengthen the product rather than commoditize it.

That repricing feels durable even if software stocks continue recovering.

Public SaaS cohort	Median revenue multiple
20% to 30% growth	5.5x
10% to 20% growth	3.1x
Under 10% growth	1.9x
Broad SaaS median in July	3.8x
SaaS peak in 2021	16.9x

Could AI actually make the strongest SaaS companies bigger?

Yes. AI could concentrate software spending around a smaller number of important platforms and make some of today's strongest SaaS companies even harder to displace.

The economics are easier to understand once we stop imagining software usage as one employee clicking around one application.

An agent can query databases, inspect tickets, update records, trigger workflows, check permissions and run security scans continuously. One agent may interact with enterprise software far more frequently than the human employee whose work it helps automate.

That is great economics for the platform if the platform charges for those interactions.

It also makes trusted context more valuable. ServiceNow currently has $29 billion of remaining performance obligations. Salesforce has more than $66 billion. Snowflake has $9 billion. These companies already sit inside thousands of large organizations and hold data or workflows that new agents need to access.

AI can therefore create two opposing effects at once.

A company might eliminate three narrow SaaS products because Claude can handle their workflows, then increase its spending on Snowflake because agent workloads create more data consumption. It might reduce support seats while paying Salesforce for billions of automated actions. It might consolidate security vendors while spending much more with CrowdStrike because thousands of autonomous agents have created a larger attack surface.

Current cybersecurity results are already suggestive. CrowdStrike's revenue grew 26% in its latest quarter and net new ARR accelerated 51%. Okta's remaining performance obligations grew 17%, with management explicitly positioning machine and agent identity as a new security problem.

So fewer human seats can coexist with more software consumption.

That possibility is one reason the “AI kills SaaS” argument has become much harder to state cleanly.

What would make the SaaSpocalypse much worse from here?

The SaaSpocalypse would become much more serious if today's pricing and procurement pressure starts turning into broad revenue contraction.

The number we would watch most closely is net revenue retention. Benchmarkit's seat-priced cohort is already at 98% median NRR. If established SaaS companies start reporting NRR materially below 100% across several categories, customers would be shrinking faster than vendors can compensate through ordinary expansion.

The second test is build versus buy. As seen above, McKinsey already finds 32% of organizations skipping at least one software purchase because they can build internally. The next step would be much more consequential: companies replacing whole systems rather than isolated products, features and internal workflows.

The third test is where enterprise context lives.

Claude can currently operate through Salesforce because Salesforce still owns useful customer data, permissions, business rules and actions. AI agents can work through Jira because Atlassian still holds the projects, history and workflows. Workday agents are valuable partly because Workday already sits on the HR and finance data.

The scary scenario for incumbents begins when companies decide that the AI layer can own enough of that context itself.

We are nowhere close to proving that at broad enterprise scale today.

If it happens, the SaaSpocalypse moves from attacking dashboards and seats to attacking the core reason major SaaS platforms exist.

So, is the “SaaSpocalypse” real?

Yes, but in a narrower and more interesting way than the name suggests: weak SaaS economics are already being disrupted, while software demand itself is still growing.

We found several changes that now look too large to dismiss as investor paranoia.

Nearly one-third of organizations in McKinsey's latest survey have already skipped at least one software purchase because they could build the functionality with coding agents. Usage pricing has overtaken seat pricing in Vertice's dataset. Benchmarkit finds 108% median NRR for usage-priced companies versus 98% for seat-priced companies. Gross retention across SaaS has weakened. Gartner estimates that agents could expose $234 billion of enterprise application spending by 2030.

Those numbers describe a genuine break with the old SaaS playbook.

The other half of the evidence is equally strong. Private B2B SaaS companies are still growing at a 22% median. The ten large software companies we checked were growing at a 23% median in their latest quarters. Snowflake just grew 35%. Datadog grew 36%. ServiceNow grew 24%. AI spending inside enterprises is exploding, and several established SaaS companies are now generating billion-dollar AI revenue streams.

The SaaSpocalypse is real for software whose value came mainly from putting a convenient interface around a reproducible workflow and charging every employee for access. AI can attack the interface, the workflow and the seat count at the same time.

Software that owns valuable data, permissions, security, transactions or deep operational context has a much stronger position. In several cases, AI is already driving more usage through those systems.

The easy SaaS model is getting crushed much faster than SaaS itself.

That distinction explains why both sides of this debate can point to convincing evidence. The February selloff saw the disruption clearly but priced it far too quickly. The recent software rebound correctly recognizes that incumbents still have powerful assets, but it would be a mistake to assume the old economics are coming back.

As of now, “SaaSpocalypse” works as a name for the destruction of the old SaaS playbook, but it badly describes what is happening to software demand.