Software is dead. Again.
At least this is the market’s point of contention. The median revenue multiple for cloud software has collapsed to 4.1x, a level not seen since the Fed spooked markets in early 2016. But here's the number that should get your attention: the median free cash flow multiple is 18.9x, about 30% below the previous decade’s low.
Let that sink in. Investors are pricing software companies lower on a cash-flow basis than at any time in the last 10 years. Not during the 2016 scare. Not during the COVID crash. Now.
The obvious explanation is AI. Vibe coding. The marginal cost of software creation is approaching zero. Those forces are real. But the consensus narrative that AI will commoditize all software into oblivion misses the actual structural shift happening beneath the surface.
The Two Broken Assumptions
To understand what's really happening, you need to understand how software companies have been valued for the past two decades.
The SaaS model was sold to investors as a "cash flow annuity." Lose money early acquiring customers, flip profitable, then print predictable cash forever. The math was elegant: sum the present value of ten years of projected cash flows, slap on a terminal value assumption, and you've got your intrinsic value.
This model rested on two critical assumptions. First, retention rates remain high and stable—because without that, your cash flow projections collapse. Second, there is a terminal value. As in, the company still exists and generates cash in year eleven and beyond.
AI has put both assumptions on trial.
The bear case writes itself: if an AI agent can rewrite a codebase or migrate data to a cheaper platform in minutes, switching costs evaporate. If retention drops from 95% to 80%, the annuity math breaks. If your software can be replaced by a "vibe-coded" internal tool, your terminal value is not a multiple of cash flows; it is zero.
Here's the key insight: AI does not have to kill revenue today to blow up valuations. It just has to widen the distribution of outcomes. Introduce plausible zero-terminal-value scenarios for enough companies, and you justify a higher discount rate across the sector. Investors are not saying these businesses are bad. They say the right tail just got thinner, and the left tail got fatter. This broader uncertainty is at the core of current market repricing, even if the surface narrative misses it.
The Narrative Violation
If the "software is dead" thesis were universally true, we'd expect carnage across the board. But look closer at the data, and you'll find what I call a "narrative violation."
The growth-adjusted revenue multiple, which divides the valuation by the growth rate, is 0.35x. That is higher than the pre-COVID average of 0.28x. In a world of scarce growth, investors pay a premium per unit of growth for the few companies that can still deliver it.
ServiceNow just reported Q4 earnings, with 98% retention. No deterioration. Workday raised guidance. Salesforce signed 3,000 paying customers for Agentforce in a single quarter.
The median is masking a brutal bifurcation. This isn't a cycle—it's a sorting event.
The Publisher Parallel
Here's what I think the market is actually pricing in, even if it can't articulate it clearly.
When the internet first arrived, it seemed like a tremendous opportunity for publishers. Suddenly, their addressable market was not just their geographic delivery area; it was the entire world. But the nature of the opportunity was the opposite. Free distribution was not exclusive. It was available to every competitor, every blogger, everyone. What helped any individual publisher was disastrous for publishers as a group.
The same dynamic is now hitting software. AI-written code is an advantage for any individual software company; they can write more code, faster, and more efficiently. But that advantage is not exclusive. Every competitor has it too. What is changing is not distribution as in media; it is the cost of the input itself: code.
AI doesn't kill software demand. It kills software scarcity.
The barrier to building a decent vertical SaaS tool is dropping from $5 million to $50,000. When supply becomes effectively infinite, prices converge toward marginal cost. For software that is approaching the cost of inference, pennies on the dollar compared to traditional SaaS subscriptions.
This is catastrophic for "middle SaaS," the thousands of companies selling what amounts to forms on top of a database. Employee leave trackers. Simple inventory tools. Basic OKR software. Competition, not substitution, is the silent killer.
Why Software Won't Fully Commoditize
Before we bury the industry entirely, three structural factors will prevent complete commoditization.
First, companies focus on core competency. For most businesses worldwide, that isn't software. There's a reason companies pay other companies for software, and that reason won't change with AI.
Second, writing the original app is just the beginning. There is maintenance, security patches, new features, and changing standards. Writing an app is a commitment to a never-ending journey, one that has nothing to do with most companies' core business.
Third, selling software is not just selling code. There is support, compliance, and integrations with other systems. Companies do not run purely open-source software for the same reason they will not rely on vibe-coded internal tools for mission-critical functions: they do not want code; they want a product.
But here is the catch: these protections apply to platforms, not to commodity point solutions. The moat exists for systems of record. It does not exist for nice-to-have horizontal SaaS.
The Build vs. Buy Renaissance
The downstream effect is a fundamental shift in how CIOs think about procurement. For fifteen years, the philosophy was buy, not build. But when your team can vibe code an internal tool in an afternoon, that math changes completely.
The new framework is not build vs. buy; it is core vs. context.
For Systems of Record like ERP, CRM, and HCM, you still buy. The risk of building your own Sarbanes-Oxley-compliant HR system is too high. But for the glue code, edge workflows, and internal utilities, why pay $10 per user per month when an AI agent can write the integration script for free?
This structurally shrinks the addressable market for mid-market SaaS while increasing the importance of platforms that host build activity.
Fighting for the Pie
Here's where the investment thesis gets uncomfortable.
For the last decade, the SaaS story was about growing the pie. Identify a business function, build an app, hire sales, run cohort analysis, and go public. The playbook was clean and infinitely repeatable.
That story is over. Businesses do not want to buy more software. If anything, they need to cut spending to allocate a budget for their own AI tokens. The growth narrative for most software companies is structurally impaired. The industry-wide re-rating is not an overreaction. It is correct.
The next decade will not be about growing the pie. It will be about fighting for it. Software companies will use AI to attack adjacent markets, justify their existence, and raise prices by expanding their scope. The model makers, such as OpenAI, Anthropic, and Google, will be the arms dealers.
Platform Advantage ≠ Platform Safety
Now here is where I will temper my own thesis. Platforms have real advantages: data gravity, governance, compliance, and trust. But two risks deserve attention.
First, platform fragmentation. If every major vendor becomes an AI control plane, enterprises may run multiple control planes, arbitrage pricing, and reduce lock-in over time. Being a platform is an advantage, not a monopoly.
Second, the identity layer is shrinking. Microsoft built its empire on Active Directory, organizing companies by identity and then monetizing per seat. But the value of identity ownership declines as the number of human identities dwindles. Per-seat licensing breaks down when agents replace workers.
Here is a subtler point: retention stability does not equal pricing power stability. Holding 98% retention is necessary but not sufficient. AI pressure may show up as slower expansion revenue, usage-based compression, or renegotiated contracts, all while retention technically holds. The old SaaS math required both high retention and consistent expansion. One without the other changes the terminal value calculation.
What Changes the Market's Mind
The path to multiple re-expansion isn't a single blowout earnings report. It's the re-establishment of confidence in terminal value.
Investors need four to six quarters of data showing that retention rates and pricing power for AI-enabled platforms are sustainable. That outcome-based pricing can capture revenue even as seat counts decline. The Service-as-a-Software model, where software absorbs both labor and IT budgets, actually works at scale.
ServiceNow's Q4 was a green shoot, not proof. The next eighteen months will determine whether platforms are true monopolies of intelligence or just temporarily advantaged incumbents.
The Sorting Has Just Begun
The days of buying a basket of cloud stocks and watching them triple are over. The index-level beta is broken.
Value concentrates where data is proprietary, where switching costs are institutional rather than technical, and where trust matters more than features. The market is right to destroy median SaaS. It may be early, but it's not wrong to question whether even platforms deserve unconditional faith.
An 18.9x FCF multiple isn't pure fear. It's the market correctly recognizing that "SaaS 1.0"—growth at all costs, infinite retention assumptions, seat-based pricing—is coming to an end.
What replaces it won't be smaller. It will be more concentrated, more platform-driven, and brutally competitive at the edges. The software industry isn't dying. But it's being sorted.
And the sorting has just begun.