In the twelve months to early February 2026, software stocks went through their largest non-recessionary drawdown in more than thirty years. J.P. Morgan's Dubravko Lakos-Bujas put the fall at 34 percent, roughly two trillion dollars of market cap wiped out from the peak, and blamed mounting concern over the disruptive impact of new LLM capabilities. This was not a panic. Investors reassessed what software is worth when the things that made it defensible stop being scarce.
Three SaaS moats are eroding at once
Three assumptions got repriced at once. Software is no longer hard to build, because anyone with technical capability can replace a SQL wrapper on a billing system in a weekend with Claude Code. Seat expansion is no longer durable, because agents are starting to do the work that used to require a human at a license. Feature and interface moats, which drove SaaS competitive dynamics for fifteen years, are commoditizing as agents render the interface itself optional. These three assumptions are the structural reasons SaaS multiples sat where they sat, and all three are eroding at the same time.
The churn data tells the same story. ChartMogul's retention report, which segments roughly 3,500 software companies into B2B SaaS, B2C SaaS, and AI-native cohorts, put net revenue retention for the AI-native cohort at 48 percent against a B2B SaaS median of 82 percent, and found AI products priced under fifty dollars a month retaining just 23 percent of gross revenue. ChartMogul's phrase for the pattern is the curse of the AI wrapper: the downside of being easy to buy is being easy to cancel.
Services-as-software is what comes next
What comes next has a name: services-as-software. Foundation Capital coined the term in early 2024. Bessemer wrote the playbook for vertical AI variants of it in September 2024. Sequoia made it the title of a March 2026 piece (Services: The New Software, by Julien Bek). The argument across all three is the same. The customer never wanted the tool, the customer wanted the work done. The 2010s SaaS playbook sold the tool because that was the only thing the technology could deliver. The 2026 playbook sells the outcome, because that is what AI delivery now makes possible.
The labor TAM behind the shift is six times the software TAM
The TAM math behind the shift is the part that surprised most public-market investors. Sequoia's framing is six dollars of services spending for every one dollar of software spending. General Catalyst's Marc Bhargava puts it at sixteen trillion in services versus one trillion in software, a ratio of fifteen times. Foundation Capital's headline number is the $4.6 trillion services opportunity their original April 2024 piece sized. The point is not the precise multiple. The point is that the market software was eating in 2015 was less than a sixth of the market AI is now positioned to eat, and the strategies built for the smaller market are not the strategies that capture the larger one.
Which SaaS is dying, and which is not
This does not mean every SaaS company is dying, and it is worth being precise about which ones are.
The defensible SaaS businesses are not getting replaced. Salesforce, ServiceNow, Snowflake, Stripe, the companies whose moat is proprietary data, regulatory positioning, deep integration into a customer's systems of record, or a network effect that compounds with use, are absorbing AI as a layer on top of their existing surface, the way they absorbed mobile in 2012 and the API economy in 2015. The SaaS CFO's framing is the right one: SaaS is separating defensible from convenience-based, and the data moats become more durable in an AI world rather than less, because agents without proprietary context are far less valuable than those with it.
The SaaS that is dying is the convenience-based middle layer. The marketing automation tool whose differentiator is a scheduling UI. The prospecting platform whose value is a database of phone numbers. The workflow point solution that automates a single step of a five-step process and charges fifty dollars a seat for the privilege. These are the AI-wrapper churn cases. These are the categories where seat-based pricing becomes incoherent the moment an agent at the customer's end can do the work without occupying a seat. SaaStr's framing for the category is uncompromising: your AI is not your moat, your AI is table stakes, and the moat has to come from somewhere else.
Where the moat comes from in a services-as-software world
The question worth taking seriously is where the moat now comes from.
In a services-as-software world, the durable competitive positions resemble the positions services firms have always held rather than the positions SaaS firms held. Vertical depth comes from doing the work in a specific industry for years, building proprietary data and pattern libraries that nobody outside the vertical has. Embedded delivery teams put engineers in the customer's environment, accountable for outcomes, and they surface the unwritten rules that nobody could specify in advance. Outcome-aligned pricing is the only structure where vendor and customer share exposure to whether the agent actually works. Network effects come from aggregating outcomes across customers in the same vertical. None of these look like the SaaS playbook.
Bessemer's vertical AI piece names this directly. Vertical AI's market capitalization, they predict, will reach at least ten times the size of legacy vertical SaaS, whose top twenty US public companies combined for roughly $300 billion in market value when the piece ran. The reason is mechanical. Vertical SaaS sold software into a software budget. Vertical AI sells outcomes into a labor budget, and the labor budget is an order of magnitude larger.
The honest version of the argument is therefore narrower than the headline. SaaS is not dying; the SaaS playbook of the 2010s is. The companies that will define enterprise software in the 2030s will not look like the SaaS companies that defined it in the 2010s, because they will be priced and sold and delivered differently, and built around a different unit of value. The next decade's category-defining company in the sales-tooling category will not be a CRM. It will be the company that sold the sales work, with the software running underneath it.
That is what the title means. SaaS as a model is dying while services-as-software, the model that replaces it, is rising, and the two are not the same business at all.