Monday, May 25, 2026 looked like a simple risk-on story.

Oil dropped more than 5 percent. Nasdaq futures led. Japan's Nikkei jumped about 3 percent. The clean headline was that optimism around a U.S.-Iran deal lifted risk assets and crushed crude.

That read was not wrong. It was just lazy.

The more useful signal was where attention kept clustering underneath the macro move. The interesting names were not only the usual AI giants. The market kept circling the less glamorous layer: storage, networking, enterprise hardware, power conversion, and the systems that make AI work after the keynote ends.

That is the second wave of the AI trade.

It is not sexier models. It is the plumbing finally getting paid.

What happened

The U.S. cash market was closed for Memorial Day on Monday, May 25, 2026. That matters because a holiday board is not the same thing as confirmed institutional sponsorship. It is a preview, not a verdict.

Still, previews can be useful when leadership starts to cluster in the right places.

The overnight setup was clearly risk-on. Global equities moved higher, oil fell sharply, and the macro tape gave traders an easy headline to chase into Tuesday's open. But the names surfacing inside that move were more revealing than the headline itself.

Dell has spent May pushing its AI Factory story harder, with fresh emphasis on enterprise deployment, liquid-cooled systems, and on-prem infrastructure. HPE has been pushing from a parallel angle, with messaging around autonomous networking, private cloud, storage, and data readiness for production AI. NetApp has kept leaning into the same lane, framing itself as intelligent data infrastructure for the AI era. Navitas is an even cleaner tell: its recent materials are not about chatbot excitement at all, but about power density, 800V architecture, and the physical demands of next-generation AI factories.

That cluster matters more than one green futures board.

When a market theme starts to mature, money usually broadens from the obvious winners into the layers that real budgets cannot avoid. That is what this setup looked like.

Why it matters

Too much AI coverage is still trapped at the demo layer. Which model is smarter. Which assistant is faster. Which launch looked the coolest. That is fine if the job is farming attention. It is weak if the job is understanding where durable spend is going.

Real AI adoption does not stop at the model. It has to run through compute, storage, networking, cooling, power delivery, data pipelines, governance, and deployment infrastructure. That is where budgets get harder to fake.

A company can tell a beautiful AI story on stage and still get exposed in production. Where does the data live? How fast can it move? What breaks first under load? How expensive is inference at scale? What powers the rack? Who controls the environment?

Those questions do not create viral headlines. They do create revenue.

That is why this matters for operators too, not just investors. The AI market is moving out of the novelty phase and deeper into the implementation phase. In the novelty phase, language wins. In the implementation phase, infrastructure wins.

The opinionated take

The next leg of AI will reward businesses that make the stack work, not businesses that merely talk about the stack well.

That should be obvious. The market still needs to relearn it on schedule.

For a while, AI beta was broad enough that adjacency did most of the work. If you touched the theme, you borrowed excitement. If you said agents, copilots, or automation with a straight face, people filled in the blanks for you.

That phase is getting weaker.

Budgets are becoming more selective. Macro pressure has not disappeared. And once a theme gets crowded and expensive, capital starts asking a better question: what is actually necessary here?

That is why the boring layer matters now.

Storage is necessary. Networking is necessary. Power architecture is necessary. Data readiness is necessary. Enterprise hardware that gives customers control over latency, cost, and security is necessary.

The model can change. The plumbing still gets paid.

That does not make every infrastructure name a winner. It does mean the market is finally looking below the headline layer for businesses attached to unavoidable work. That is a healthier phase of the trade, but it is also a more demanding one.

The winners from here are less likely to be the loudest storytellers. They are more likely to be the companies solving hard, physical, unglamorous constraints inside real deployment environments.

That is where the money usually gets serious.

Practical takeaway

If you are an investor, stop treating AI like one giant basket. Break it into cleaner layers:

  • model companies
  • app-layer wrappers
  • workflow and governance platforms
  • data infrastructure
  • storage and networking
  • power and physical buildout

Those layers do not deserve the same assumptions, and they definitely do not deserve the same valuation logic.

If you are a founder or operator, the takeaway is even cleaner. Stop trying to sound AI-native and start trying to be operationally necessary.

Ask harder questions:

  • Does this reduce deployment friction?
  • Does this lower inference cost?
  • Does this improve control?
  • Does this make data more usable?
  • Does this make production AI more reliable?

If the answer is yes, you are closer to the budget line that survives. If the answer is no, you may still have a demo, but you do not necessarily have a durable business.

The real read

Monday's setup did not prove the AI trade is easy again. It proved something better.

The market keeps sending attention and capital toward the layers that make AI physically and operationally real. That is a more useful signal than another model-comparison headline.

The first wave of AI excitement was about intelligence.

The second wave is about infrastructure.

And infrastructure, as usual, looks boring right before it matters most.