The AI Trade Is Learning to Rent Applications Again

Last week the AI trade looked simple again: memory chips ripped, Nvidia was still the market's metronome, and every fresh dollar wanted to sit as close to the data center as possible.

This week the more interesting move was quieter. Software stocks started to bounce, and Workday helped by posting a quarter that was good enough to calm the market's favorite fear: that AI would flatten traditional enterprise applications faster than incumbents could adapt.

That is the wrong way to frame the next leg of the trade.

The question is not whether AI kills software. The question is what companies rent after they buy the intelligence.

A CFO does not approve a giant model budget so the finance team can admire a smarter chatbot. A CHRO does not renew core HR systems because employees enjoy better prompts. The rent check gets written when someone can turn raw model capability into payroll runs, hiring approvals, expense controls, procurement steps, and month-end close work that still has to happen on Tuesday morning.

That is why the software rebound matters.

Reuters reported on May 19 that investors were starting to separate software companies that look exposed from those that remain deeply embedded in enterprise workflows. Workday, ServiceNow, and Salesforce all rose that day even as chip momentum cooled. The market was not suddenly feeling sentimental about SaaS. It was beginning to admit that after the infrastructure spend comes the operating layer.

Workday's own quarter made that easier to believe.

Reuters reported on May 22 that Workday's subscription revenue rose 14.3% to $2.35 billion, with net new business driving 40% of that growth. The company beat revenue and profit expectations, and the stock jumped nearly 12% premarket. Workday did not solve every software valuation problem in one earnings call. It did something more useful: it showed that customers are still paying real money to keep core systems in place while AI features get folded into them.

Investors have spent most of 2026 acting as if enterprise software had only two futures.

• One future said the model companies would commoditize the application layer.
• The other said incumbents would slap "AI" on existing products and keep the old economics forever.

Neither view matches how large companies actually buy tools.

Enterprise budgets usually move in sequence. First comes the panic spend, where boards approve infrastructure, compute, and data projects because nobody wants to be caught behind. Then comes the slower spend, where managers ask a more boring question: which vendor will reduce the amount of human coordination wrapped around everyday decisions?

That second budget is where software can recover.

Not all of it. Reuters' sector piece made that clear too. Markets are drawing a line between companies tied to per-seat subscription growth and companies that sit closer to the center of AI deployment. That distinction matters because AI can absolutely compress some software categories. If a product is basically interface labor with weak switching costs, the model may eat it.

But if a product is the place where tasks get approved, routed, reconciled, and recorded, AI may make that product more rented, not less.

That sounds backward until you picture the actual office scene.

A controller still has to sign off on the close. A recruiter still has to move a candidate from interview to offer. A procurement team still needs spend rules, audit trails, and who-approved-what clarity. AI can accelerate the work around those steps, but it does not eliminate the value of the place where the step becomes official.

This is why the software rebound should worry anyone still treating the AI trade as a pure hardware story.

The first phase of the boom rewarded whoever sold scarcity: GPUs, memory, networking, power, land, cooling. The second phase is likely to reward vendors that can convert expensive intelligence into cheaper operating motion inside the enterprise. That is a different kind of monetization. It is less cinematic than a trillion-dollar chip rally, but it can be stickier because it gets buried inside recurring process.

There is also a market consequence here.

If the infrastructure names keep absorbing all of the multiple expansion while the application layer quietly proves it can defend retention and widen wallet share, investors may be underestimating how much AI spending eventually migrates from capex stories to opex stories. The same CFO who approved more data-center capacity still needs a reason that labor, errors, and cycle times actually improve. If the hardware bill is the first check, the workflow bill is the second.

That is why software's bounce deserves more respect than a short-covering headline.

It may be the market's first admission that once companies buy intelligence, they still need to rent coordination. And coordination, annoyingly enough, is where the real money tends to stay.