Jordi Visser's Contrarian AI Bet: Why He's Buying Transformers, Not Software

The 22V Research strategist is telling clients to forget AI growth stocks. The real money, he argues, is in the boring industrial companies that will power the next phase.

Jordi Visser has a message for investors chasing the next Nvidia: you're looking in the wrong place.

The head of AI Macro Nexus Research at 22V Research has been telling clients to rotate out of AI software names and into something far less sexy—electrical transformers, cooling systems, and power infrastructure. It's a contrarian call that's getting attention precisely because it sounds so boring.

"Everyone's still fighting the last war," Visser wrote in his 2026 outlook. "They're buying the brain builders when they should be buying the body builders."

His thesis, which he's been developing publicly on X and in client notes, centers on a simple observation: AI is about to leave the cloud.

The Physical World Upgrade

For the past three years, AI investment has been dominated by what Visser calls the "brain building" phase. Companies trained ever-larger models, requiring ever-more powerful chips and ever-bigger data centers. Nvidia captured most of the value. A handful of hyperscalers captured the rest.

But training is reaching a plateau. The models are built. The next phase is deployment—getting AI out of centralized clouds and into the physical world. Into factories. Into cars. Into robots. Into edge devices that can see, move, and act.

This phase, Visser argues, requires a completely different set of investments. Not more H100s. More transformers. Not more software. More copper.

"We spent 2023 and 2024 teaching AI to think," he told clients. "We're going to spend the next twenty years teaching it to see, move, and build. That's a hardware story, not a software story."

The PMI Signal

Visser's framework relies heavily on the Purchasing Managers' Index—a diffusion measure that tracks whether more firms are expanding or contracting in a given month. He argues that PMI has been suppressed during the AI training phase because the benefits were concentrated in a handful of tech giants. The broader industrial economy didn't feel it.

That's about to change.

His timeline is specific. The 2023-2024 period was AI utility buildout—training the models. 2025 is grid reliability and cost compression—making the infrastructure stable and affordable. 2026 is enterprise and edge deployment—when AI spreads from tech companies to everyone else.

"PMI improves not when AI is built, but when AI spreads," he wrote. "We're moving from brain builders to economy-wide beneficiaries."

The evidence he points to: semiconductor capital expenditure is broadening beyond the hyperscalers. Industrial automation orders are accelerating. Power infrastructure backlogs are at multi-decade highs.

The Bottleneck Map

Rather than naming specific stocks, Visser identifies "bottleneck asset categories"—areas where supply constraints will create pricing power as AI deployment accelerates.

The list reads like a syllabus for an electrical engineering course:

Power transformers are at the top. Lead times have stretched to three to six years. Shortages could hit 30% by the end of 2026. Every data center, every factory adding AI capabilities, every grid upgrade requires transformers. There simply aren't enough.

Thermal management comes next. AI inference generates heat. The transition from air cooling to liquid cooling is just beginning. Vertiv, nVent, and a handful of others dominate the space.

Advanced packaging is third. TSMC's CoWoS capacity—the technology that connects AI chips—remains severely constrained despite massive expansion plans.

Networking equipment rounds out the list. AI data centers require dramatically more copper than traditional facilities. The suppliers of high-speed connectors and cables face demand they can't meet.

The Crypto Connection

Visser's framework includes an unexpected prediction: this is the year crypto dominates growth investing.

The connection isn't ideological—it's statistical. Historically, Bitcoin and altcoins perform best when PMIs are rising. If Visser's thesis is right and 2026 brings a PMI inflection, crypto should benefit.

"Watch the 50-day moving average for BTC and ETH," he advised followers. "That's your signal."

It's a bold call given crypto's volatility, but it's consistent with his broader framework. When industrial activity expands, risk appetite increases. When risk appetite increases, speculative assets rally. He's betting on both.

The Liquidity Trap

Visser's bullishness comes with a significant warning.

If he's right—if investors do rotate en masse from software to hardware, from cloud to physical—the transition could be violent. There's a mismatch between the liquidity of software stocks and industrial stocks. You can sell a billion dollars of Salesforce in an afternoon. Selling a billion dollars of Schneider Electric takes longer.

"I worry about a violent liquidity event where the door to the theater is simply too small for the crowd trying to exit," he wrote. The software selloff earlier this week may have been a preview.

The other risk is timing. Visser's thesis requires PMIs to actually inflect. If deployment stalls—because of power constraints, regulatory friction, or simply slower enterprise adoption—the industrial trade could underperform for longer than investors can wait.

The Counter-Narrative

Not everyone is buying the physical world upgrade story.

Skeptics point out that Visser has been calling for this rotation for over a year. Industrial stocks have traded sideways while AI software names kept rallying. Being early is indistinguishable from being wrong until suddenly it isn't.

Others question whether the AI deployment phase will be as capital-intensive as Visser expects. Cloud inference is getting cheaper. Edge devices are getting more efficient. The infrastructure demands could be lower than projected.

And there's the competitive question. Even if transformers and cooling systems become scarce, that scarcity attracts capital. New capacity comes online. Bottlenecks ease. Pricing power fades.

Visser's response: look at the lead times. Power infrastructure has multi-year build cycles. You can't spin up a transformer factory the way you can spin up a software company. The bottlenecks aren't going away quickly.

What It Means for Investors

The practical implications depend on your time horizon.

For tactical traders, Visser's framework suggests watching PMI data closely over the next few months. An inflection—especially above 50 in the U.S. and Europe—would validate the thesis and potentially trigger the rotation he's predicting.

For longer-term investors, the message is subtler. The AI trade isn't over, but it may be changing shape. The winners of the next phase might not look like the winners of the last phase. A portfolio that's all Nvidia and Microsoft might underperform a portfolio that's added some boring industrial names.

For the truly contrarian, there's a version of this trade that's even more aggressive: short the software stocks that are most exposed to AI disruption while going long the infrastructure names that benefit from deployment. It's essentially betting that this week's selloff was the beginning, not the end.

Visser isn't giving specific stock picks publicly. But his framework points in a clear direction: away from the cloud and toward the grid. Away from the brain and toward the body. Away from software and toward stuff.

In a market that's spent three years obsessing over artificial intelligence, he's telling investors to focus on actual infrastructure. It might be the most contrarian AI take of 2026.

---