Stephanie ColeWall Street Is Wrong About AI Killing Software Companies
February 28, 2026
I was in a car on the way back from a lunch thing - the driver was taking the long route, which I only know because my assistant mentioned it later - and I was reading about the SaaSpocalypse on my phone. That's the actual word people are using. SaaSpocalypse. I had someone pull up more context for me when I got back to the office, and I've been thinking about it since. Because I think Wall Street has gotten this one genuinely, embarrassingly wrong. Not slightly wrong. Structurally wrong.
Here is what happened. Anthropic released a productivity tool for in-house lawyers, and shares of legal software and publishing firms went tumbling. That single launch - Claude Cowork - triggered a single-day $285 billion wipeout, with Salesforce and Workday each down over 40% in the prior 12 months. A massive valuation reset wiped out nearly $1 trillion in market value, sending several industry giants to 52-week lows. The trading desks lost their minds. Sentiment went from bearish to doomsday, with traders dumping shares across the industry. One Jefferies trader called it the "SaaSpocalypse" - an apocalypse for software-as-a-service stocks.
And the fear has a very specific shape. Investors are moving away from the "per-seat" licensing model - the bedrock of SaaS for two decades - fearing that autonomous AI agents will drastically reduce the number of human employees needed. The concerns are brewing in private equity as well, with firms including Arcmont Asset Management and Hayfin Capital Management hiring consultants to check their portfolios for businesses that could be vulnerable. Apollo cut its direct lending funds' software exposure almost by half in 2025, from about 20% at the start of the year.
It looks like panic. It feels like panic. And I think it basically is panic.
The Argument Wall Street Is Actually Making
To be fair to the doomsayers, there is a version of this argument that's coherent. It's not that AI replaces the software directly. It's that AI reduces the headcount that uses the software. If 10 AI agents can do the work of 100 sales reps, you don't need 100 Salesforce seats anymore. You need 10. That's a 90% reduction in seat revenue for the same work output. That's a real argument. I'm not going to pretend it isn't.
And the Mistral CEO threw fuel on the fire recently, telling CNBC that more than 50% of enterprises' current software could be replaced by AI. He described customers building fully custom applications to run procurement or supply chain workflows in a couple of days - workflows that five years ago would have required a dedicated vertical SaaS product. Retool released data showing 35% of respondents have already replaced the functionality of at least one SaaS tool with a custom build, and 78% expect to build more of their own tools in 2026.
So there are real signals here. I'm not dismissing the signals. What I'm dismissing is the conclusion Wall Street drew from them, which is that enterprise software companies are about to become the new Blockbuster.
The Part Wall Street Got Very Wrong
Here is the thing about panics: they don't distinguish between things that are similar. They destroy everything in a category. Just 67% of software companies in the S&P 500 have beaten revenue expectations this earnings season, compared with 83% for the overall tech sector. That's not nothing. But while all software stocks have beaten earnings expectations, that's mattered little in the face of concerns about long-term prospects. Companies beating earnings got sold off anyway. That's not a rational market response to fundamentals. That's fear.
The software companies most vulnerable to AI disruption are a very specific type: companies with narrow, task-specific functionality that AI can replicate - like legal document review, basic data analytics, or simple workflow automation - and high per-seat pricing models exposed to headcount reduction. That's not the whole software industry. That's a slice of it. Deeply embedded systems of record like CRM and ERP platforms face less immediate risk due to high switching costs and deep integration.
Tory was doing a whole thing last week about how he's simplifying his life - getting rid of things that don't add value, making space for what matters. He lost his car around the same time, I think unrelated. I nodded along. But it made me think about this: the software companies that matter, the ones actually embedded in how enterprises run, aren't getting simplified out of existence. They're load-bearing walls, not decorative ones.
To date, installed bases have largely held up, with average gross retention still around 90% or better. Customers are not leaving. They're just not adding seats as fast. Customers say they would prefer to buy AI-enabled solutions from their incumbent vendors. They trust them. They know they are secure. They believe they will be around for the long term. That's not the behavior of an industry being replaced. That's the behavior of an industry being asked to evolve.
The Real Story Is About Budget, Not Replacement
The most honest take I've read on this whole situation came from SaaStr founder Jason Lemkin, and I think he has it exactly right. Lemkin argues the crash isn't AI killing SaaS at all - it's "the market finally pricing in the deceleration that started in 2021." Growth was already slowing before AI. The AI narrative just gave Wall Street a story to attach to the numbers.
Public SaaS growth rates have declined every single quarter since the 2021 peak. Every. Single. Quarter. This isn't new. The AI crash narrative just gave the market permission to finally re-rate what the numbers have been screaming for three years.
What's actually happening to software budgets is more subtle and more interesting than "AI is killing SaaS." AI isn't eating the product. It's eating the budget. The Big Five hyperscalers plan to spend between $660 billion and $690 billion on infrastructure in 2026, nearly doubling 2025 levels. Approximately 75% of this spend targets AI infrastructure. This massive capital redirection is directly competing with enterprise software for corporate IT budgets.
Chris asked me the other day whether I thought companies would just... stop buying software. He had this slightly worried look, the way he gets when he's actually thought something through. I told him no - and I meant it. The question isn't whether enterprises spend on software. It's which software, and in what proportion to their AI infrastructure spending. That's a very different problem from extinction.
We write about this dynamic a lot around here - the way SaaS stocks tank while renewal quotes go up is its own kind of absurdity, and HSBC coined the SaaSpocalypse term well before the actual selling started. The vocabulary was there. The market just needed a trigger.
What Actually Gets Replaced - And What Doesn't
Let me be specific, because specificity is what the doomsday narrative lacks.
"Sticky" software - ERP and CRM systems that organizations have relied on for years and invested tens or hundreds of millions of dollars into - will not be discarded overnight. Nobody is dismantling their Salesforce instance because Claude Cowork launched. The switching costs are real. The institutional memory built into those systems is real. Nobody is building a homegrown CRM in Replit to replace their Salesforce instance.
What is getting replaced - or at minimum, seriously pressured - is the commodity middle layer. Most horizontal SaaS companies function as expensive intermediaries between a human user and a structured database. The narrow point solutions. The tools that do one thing and charge $40 per seat per month for it. The legal research tools, the basic analytics dashboards, the single-workflow automation products. The best-of-breed era of buying specialized point solutions is over. Every enterprise wants fewer vendors, not more.
Even the Mistral CEO, who made the headlines by saying half of enterprise software could be replaced, acknowledged that software focused on systems of records "are not going to change." That's a significant carve-out. That's most of the market cap in enterprise software. The thing causing panic affects a subset, and Wall Street sold the whole category.
I had someone set up one of the AI workflow tools our team uses - I don't know which one, honestly. She said it took most of the afternoon. When I asked if that was long, she looked at me in a way that suggested I'd said something revealing. The point is: the implementation friction is real. Only 19% of respondents in a recent survey described their organizations as having advanced AI automation maturity. Enterprises aren't replacing software at the speed the panic implies. They're experimenting. There's a difference.
The Pricing Model Is What's Actually Dying
Here's where I'll give the bears partial credit. They're not wrong that something is dying. They're just wrong about what it is. It's not the software companies. It's the per-seat model.
Companies like Salesforce and Adobe are being forced to pivot from traditional per-seat subscriptions to usage-based pricing models just to keep pace. About 70% of software providers now admit that the cost of delivering AI features is eating into their profitability. The era of infinite SaaS margins is being challenged by the high cost of GPU compute. That's a business model under pressure. That's not the same as a business under pressure.
The companies that figure out outcome-based pricing - charging for what gets done rather than who logs in - are going to come out of this in a stronger position than they entered it. Cracking this challenge will position some as the leaders of the next cycle. That's the actual story. Not annihilation. Transformation.
Derek, who I am fairly sure has never thought about enterprise software pricing in his life, told me the other day that he thinks the Disney Star Wars sequels were "structurally superior" to the originals because they had better production values. I don't have strong feelings about this. But I did think: production values aren't the product. And for enterprise software right now, the pricing model isn't the product either. The underlying capability still matters. The question is just how they charge for it.
Jensen Huang Said It Plainly
I don't often agree with things that were said at chip company press conferences, but Nvidia's Jensen Huang pushed back directly on the doomsday narrative, stating that "this notion that the software industry is in decline and being replaced by AI" is wrong - that AI augments software rather than replacing it. That's the correct framing.
The winners will be those who combine the agility of AI agents with the reliability of SaaS to deliver measurable business value. SaaS brings the workflows, governance and guardrails that enterprises demand while AI agents extend productivity and speed. One without the other falls short - but together they set the new standard for enterprise software.
The bears are modeling a world where AI replaces software. The world that's actually coming is one where AI runs on top of software, using software as the system of record and governance layer. That's not a world where software companies die. It's a world where the bad ones get culled and the good ones get more important. We've been tracking the messier version of this question - what vendors actually say when you ask them if AI will kill their product - and the honest answer is that nobody really knows yet, but the confident predictions of total replacement don't survive contact with how enterprises actually operate.
Which is also, if I'm being honest, why I think the more interesting question is what AI agents are doing inside software right now - not whether they're replacing it in theory. Theory is what Wall Street trades on. Reality is what the rest of us have to work with.
What This Means If You're Running a Business
I'll say what I actually think, which is that none of this is as alarming as the headlines suggest if you're on the buying side. The leverage has shifted to buyers. With SaaS companies desperate to show growth, you can negotiate for usage-based or outcome-based pricing models that align costs with actual value delivered. This is a good moment to renegotiate. We wrote about the awkward renewal dynamics that emerge when your vendor just had a bad earnings call - and right now, a lot of vendors have had very bad earnings calls.
The software that's genuinely load-bearing in your business - the CRM, the ERP, the compliance systems - is not going anywhere. Enterprises will face significant hurdles if they plan to rip and replace deeply integrated SaaS applications with AI-native tools. The commodity point solutions you have six of because someone bought them at different times without checking? Those are worth auditing. The average enterprise runs 130+ SaaS applications, and many companies find 20-30% redundancy when they look closely. That consolidation was overdue before AI arrived. AI is just making it more urgent.
My travel agents - I have two, one handles Europe and one handles everything else, it's just more efficient - would probably tell me this kind of thing shakes out differently in practice than it does on paper. Which is a sentiment that applies pretty cleanly to the SaaSpocalypse narrative. On paper, AI kills software companies. In practice, it's more complicated. Enterprises don't move fast. Switching costs are real. Trust compounds over years. And the companies sitting at the center of how large organizations actually function are not going to be replaced by a product that launched in February.
Wall Street panicked. The selling was real. Analysts at JPMorgan and Goldman Sachs now say the selloff has gone too far. Software price-to-sales ratios that compressed from 9x to 6x are levels not seen since the mid-2010s - which is to say, since the last time people thought they'd figured out what software was worth.
They hadn't figured it out then. They haven't figured it out now. But the companies that figure out how to price for outcomes instead of seats, and that build AI into their core workflows instead of bolting it on as a feature, are not dying. They're becoming the infrastructure of whatever comes next. That's not a eulogy. That's an upgrade cycle with worse PR.