Wall Street has a new favorite "under-the-radar" AI play, and it smells like industrial gas.
Analysts are tripping over themselves to crown Linde as the silent king of the data center. The narrative is tidy: AI chips run hot, liquid cooling is the only solution, and Linde owns the gases and engineering to make it happen. It sounds like a guaranteed win. It feels safe.
It is dangerously oversimplified.
Investors are piling into Linde because they think they’re buying a utility-like monopoly on data center thermal management. They believe they are buying "infrastructure." In reality, they are betting on a massive, unproven shift in hardware architecture that could just as easily leave Linde holding a bag of expensive, niche engineering projects while the hyperscalers find cheaper ways to move heat.
The Liquid Cooling Delusion
The bull case for Linde hinges on the transition from air cooling to liquid cooling. The logic is that as GPUs like the Blackwell B200 push power densities past 1,000 watts per chip, fans won't cut it anymore. Enter Linde, with its liquid nitrogen, specialized coolants, and heat exchangers.
But here is what the "quiet play" crowd ignores: Efficiency is the enemy of the supplier.
Hyperscalers—Amazon, Google, Microsoft—are not in the business of paying Linde a premium for proprietary cooling fluids forever. They are obsessed with PUE (Power Usage Effectiveness). Right now, they are experimenting with rear-door heat exchangers and direct-to-chip liquid cooling. Linde wants you to believe this requires their high-margin chemical expertise.
History suggests otherwise. Data center cooling tends toward the "good enough" and the "cheap." If a hyperscaler can use a closed-loop water system with basic rust inhibitors and standard chillers, they will. They don't want exotic gases. They want a commodity. Linde’s high-tech solutions are currently a luxury, not a necessity.
I’ve watched companies burn through nine-figure CAPEX budgets trying to implement "bespoke" cooling solutions, only to realize that the maintenance tail and the cost of the fluids make the ROI impossible. Linde is selling the Ferrari of cooling to a market that ultimately wants a fleet of electric buses.
The Margin Trap: Engineering vs. Gas
Linde is, at its heart, a gas company. Its valuation is traditionally supported by long-term take-or-pay contracts in the industrial and healthcare sectors. These are stable, boring, and beautiful.
The data center "play" forces Linde out of its comfort zone and into the world of complex, custom engineering.
When you move from selling bulk nitrogen to designing custom thermal management systems for a liquid-cooled cluster, your risk profile changes. You aren't just a utility anymore; you are a hardware consultant.
- Engineering costs eat margins.
- Custom builds don't scale.
- Liability increases.
If a Linde-designed cooling system fails and fries a $100 million AI cluster, who do you think the hyperscaler's lawyers are calling first? The market is pricing Linde like it gets all the upside of the AI boom with none of the hardware risk. That is a fundamental miscalculation of how industrial contracts work in the tech sector.
The "Silent" Competition Is Louder Than You Think
The competitor article claims Linde is winning "quietly." That’s code for "the market hasn't noticed the competition yet."
Linde isn't competing against other gas companies in the data center. It’s competing against Vertiv, Eaton, and Schneider Electric. These companies have been living in the server rack for decades. They already have the relationships with the site managers. They have the service footprints.
Linde is trying to break into the data center from the outside in—starting at the gas tank and moving toward the chip. The incumbents are moving from the chip out to the power grid. Who do you think has the advantage?
Think about the physical reality of a data center. Space is at a premium. Linde’s traditional model involves massive on-site storage or pipeline delivery. Data centers in urban environments or tightly packed campuses don't have room for the "Linde footprint." The engineering required to shrink industrial-scale gas tech down to a modular data center level is immense, and there is no guarantee Linde wins that race.
The Hydrogen Distraction
If the cooling narrative doesn't hook you, the Linde bulls will point toward hydrogen. They argue that as data centers face pressure to go green, Linde’s hydrogen production and carbon capture tech will power the next generation of "clean" AI.
This is a beautiful thought experiment with zero basis in current economic reality.
$$Cost_{H2} > Cost_{Grid} + Carbon_Tax$$
Until that equation flips, hydrogen is a PR stunt for data centers, not a power source. The round-trip efficiency of using electricity to create hydrogen, store it, transport it, and then turn it back into electricity via a fuel cell is abysmal compared to just using a battery.
Linde is a leader in hydrogen, yes. But selling hydrogen to a data center is like selling a gold-plated shovel to a ditch digger. It works, but it’s the most expensive way possible to get the job done. Investors are paying a "green premium" for Linde today for revenue that won't materialize for at least a decade, if ever.
Why the "Safe Play" Is the Riskiest One
People buy Linde because they are afraid of the volatility of Nvidia. They want "picks and shovels."
The problem is that the "shovels" in this metaphor are actually highly specialized, experimental thermal rigs that might be obsolete by the time the concrete dries on the next data center.
The real risk isn't that AI fails. The risk is that AI succeeds so well that it optimizes its own cooling. We are already seeing "immersion cooling" startups that use synthetic oils requiring almost no active gas management. If that tech wins, Linde’s specialized gas delivery systems become irrelevant.
You are betting on a middleman in a sector that hates middlemen.
The Institutional Inertia
Why does the "quiet play" narrative persist? Because it’s easy to sell.
Institutional investors love Linde’s balance sheet. They love the dividend. They love the "moat." But a moat is only useful if the enemy is trying to cross it. In the data center space, the tech is simply flying over the moat.
Linde is a fantastic industrial company. It is a mediocre AI company. By trying to dress it up as a high-growth tech play, analysts are setting investors up for a valuation correction. When the "AI revenue" fails to hit the hockey-stick projections because hyperscalers opted for cheaper, standard water-cooling loops, the stock will be re-rated back to what it is: a slow-growth industrial giant.
Stop looking for the "hidden" play. If it were truly the secret to the AI kingdom, the guys actually building the chips wouldn't be trying to figure out how to do it without the gas company.
Linde isn't the secret winner of the data center race. It’s the expensive backup plan for a problem that the industry is already solving with cheaper tools.
Sell the narrative. Buy the reality.
