For the better part of the last decade, software was where investors went to buy certainty.

Recurring revenue.
High gross margins.
Asset-light business models.
Mission-critical products.
And, for a long time, a market willing to pay almost any price for growth dressed up as inevitability.

That is what makes the selloff we are seeing now so important.

On Thursday, U.S. software stocks got hit again as fears about AI disruption resurfaced across the sector. Reuters reported that the S&P 500 Software & Services Index fell another 2.6% on the day and is now down 25.5% in 2026. Cloudflare, Okta, CrowdStrike, SentinelOne, Adobe, Salesforce, Intuit, and others all sold off, while Zscaler was hit even harder after a downgrade. The move was tied to fresh anxiety surrounding Anthropic’s newly unveiled Claude Mythos model and what it could mean for the future economics of traditional software businesses.

And that is the part investors need to understand.

This is not just another routine growth-stock wobble.
This is not a garden-variety pullback caused by a single bad earnings report.
And it is not merely a case of traders taking profits after a rally.

This is the market trying to answer a much bigger question:

What happens when the core value proposition of software gets challenged by the very technology that was supposed to justify higher multiples in the first place?

For two years, many software companies benefited from a simple narrative.

Add an AI assistant.
Mention copilots on the conference call.
Show a slide about productivity gains.
Tell investors you are embedding generative AI across the platform.

That was often enough.

But markets mature. Narratives harden. Then they get stress-tested.

What changed in 2026 is that investors have started separating software companies into three very different buckets.

First, there are companies that may truly become more valuable in an AI-first world because they own infrastructure, data, distribution, or a deeply embedded workflow.

Second, there are companies that can probably survive, but only if they meaningfully re-architect their products around AI instead of stapling on a few features and calling it innovation.

Third, there are companies whose margins, pricing power, or even relevance may be at risk if frontier AI models begin doing more of the work customers once paid those vendors to do. That dividing line has become a central theme in this year’s selloff.

That is why the latest round of fear hit so hard.

Anthropic’s Claude Mythos was not just interpreted as “another model release.” It was interpreted as another reminder that AI capabilities are improving faster than the market had expected, and that some parts of the software stack could be pressured sooner than investors thought. Reuters noted that the renewed selling came from exactly that concern: AI tools are becoming capable enough to automate a widening range of human and software tasks, raising hard questions about the durability of existing business models.

In plain English, Wall Street is starting to ask a brutal question:

If the software layer can be abstracted, compressed, or bypassed by AI, what exactly deserves a premium multiple now?

Plenty of investors still want to dismiss this as an overreaction.

That is understandable. After all, software has been here before. Every few years the market decides some new platform shift will wipe out incumbents. Usually, reality lands somewhere in the middle. The winners adapt, the weaker players get exposed, and the entire sector eventually marches higher again.

That may still happen here.

But this selloff feels different for one reason.

The threat is no longer theoretical.

Back in early February, Reuters described the earlier software washout as “software-mageddon,” noting that the S&P software and services index had dropped 13% in a single week and that the sector was suffering its worst relative three-month stretch since the dot-com era. Investors were already debating whether AI represented a cyclical valuation reset or a more structural threat to software economics. That debate has not gone away. It has intensified.

And now the market has fresh evidence that the pressure is not isolated to one or two names.

This week’s move tells us investors are increasingly unwilling to pay up for software simply because it is software. The old “high gross margin plus subscription revenue equals premium multiple” formula is breaking down. In its place, the market is asking tougher questions:

Can this company defend its workflow?
Can it own proprietary data?
Can it charge more because of AI?
Can it lower customer acquisition costs?
Can it deepen lock-in?
Can it become more essential, not less, as AI gets better?

If the answer is unclear, the stock gets punished.

Nowhere is this tension clearer than cybersecurity.

On one hand, this should be one of the great AI beneficiary groups. Security teams are overloaded, threat surfaces are expanding, and AI should help automate detection, triage, vulnerability scanning, and response. That is the bullish case.

On the other hand, the same advances that make AI useful for defenders can also commoditize parts of the security stack or force vendors into a much more expensive arms race. That is the bearish case.

This week alone, investors swung violently between those two views.

Earlier, Anthropic’s Project Glasswing gave cybersecurity bulls some hope. Barron’s reported that Anthropic brought in firms including CrowdStrike and Palo Alto Networks as partners, and analysts from Oppenheimer and J.P. Morgan argued this suggested cybersecurity vendors could remain central to AI adoption rather than getting displaced by it.

Then the mood flipped again.

By Thursday, CrowdStrike, Palo Alto, Zscaler, SailPoint, and Fortinet were falling sharply, and Barron’s noted that analysts were increasingly focused on which companies would actually rebuild their products for the AI era, rather than simply layering AI onto existing offerings. William Blair analyst Jonathan Ho said the winners would likely be those that re-architect around AI instead of just adding features.

That is the whole software debate right there.

AI does not automatically destroy these businesses.
But it does raise the bar dramatically.

And when the bar rises, public market investors do what they always do: they cut the multiples first and wait for proof later.

One of the more telling details in this selloff is what did not happen.

There was no catastrophic earnings miss that suddenly shattered confidence across the sector today.
No accounting scandal.
No major recession headline targeted specifically at enterprise software.
No single company blowing up badly enough to explain the breadth of the move.

That matters.

Because when a sector sells off without one clear smoking gun, what you are usually looking at is a repricing of belief itself. Reuters described the move as sentiment-driven, rooted in fear that rapid AI advancement could undermine traditional software models. That kind of shift is harder to fix because it does not get solved with one decent quarter.

A company can beat earnings and still see its stock go down if investors believe the business is less valuable three years from now.
A company can guide higher and still lose multiple points if the market thinks its moat is narrowing.
A company can talk endlessly about AI opportunity and still get sold if investors suspect that opportunity belongs more to the model providers and infrastructure owners than to the application vendor.

That is the hangover now.

The market spent years rewarding software stocks for the promise of digital transformation.
Now it is asking which of those businesses get stronger in an AI-native economy and which were only temporarily protected by the limitations of the old one.

This is also why parts of tech are behaving so differently from the software group.

Even as software has struggled, investor appetite has not disappeared from the AI trade. It has rotated.

Reuters reported in February that some investors were already favoring names like Microsoft and Oracle relative to pure application software. Since then, Oracle has helped validate that distinction: its shares rallied in March after a strong revenue forecast tied to AI infrastructure demand eased concerns over the scale of its spending plans. Reuters also reported this week that financing discussions continue around a massive Oracle data-center project in Michigan, underscoring how large the infrastructure buildout has become.

At the same time, Reuters cited Bridgewater analysis showing Alphabet, Amazon, Meta, and Microsoft are expected to collectively invest about $650 billion in AI-related infrastructure in 2026, up sharply from the prior year. Barron’s separately highlighted renewed enthusiasm for the broader AI buildout and BofA’s estimate that hyperscaler AI spending could reach $750 billion this year.

This is the key distinction investors should keep in mind.

The market may be souring on some software names, but it has not turned bearish on AI itself.

Quite the opposite.

Capital is still pouring into the picks-and-shovels layer: chips, cloud capacity, optical networking, data centers, power, and the platforms that enable enterprises to deploy AI at scale. Barron’s pointed to names like Corning as examples of legacy tech businesses benefiting from the AI infrastructure boom, while other reports showed chip stocks advancing even on days when software lagged badly.

In other words, the market is not saying, “AI was overhyped.”
It is saying, “We may have owned the wrong beneficiaries.”

That is a very different message.

Markets are crude but efficient storytellers.

Right now, the story they are telling is that software margins may not be as safe as once believed.

If AI reduces the labor required to build code, configure workflows, write content, analyze data, or handle support, then some software categories could become less differentiated. If customers can consolidate tools, pay less per seat, or rely more on AI-native platforms instead of traditional SaaS subscriptions, then pricing power comes under pressure. If implementation gets easier, switching costs may decline in certain categories. And if the most valuable intelligence increasingly sits inside foundation models or the infrastructure that serves them, some application vendors risk becoming middlemen in their own markets.

That does not mean every beaten-down software company is doomed.

Far from it.

Some of the best opportunities in the market may eventually emerge from this wreckage. Reuters noted in February that bargain hunters were already beginning to look at names like ServiceNow and Microsoft, even as others warned that valuations still had not fully adjusted to the AI threat.

That is usually how durable bottoms form.

First comes denial.
Then panic.
Then indiscriminate selling.
Then a period where good assets and bad assets get thrown together.
Only after that does the market begin rewarding real differentiation again.

We may be somewhere between stages three and four.

The next leg for software stocks will probably not be decided by slogans about AI. It will be decided by evidence.

Can companies show that AI is driving higher revenue per customer?
Can they prove that AI features are sticky, not just promotional?
Can they improve margins as they automate more of their own workflows?
Can they show that customers are standardizing on their platforms rather than questioning the need for them?

That is what the market needs to see.

And until it sees it, investors should assume volatility remains high.

For traders, that means rallies may be sharp but fragile.
For long-term investors, it means selectivity matters more than ever.
For everyone else, it means this is no time to confuse “down a lot” with “cheap.”

Because sometimes a stock is down because sentiment got too pessimistic.
And sometimes it is down because the market is beginning to understand the business more clearly than management does.

That is the danger in software right now.

The old premium is gone.
The burden of proof is back.
And until software companies can show they are beneficiaries of AI, not victims of it, this group is likely to keep trading like a sector on trial.