I keep hearing the same line in conversations lately: “I want AI exposure.”

And almost every time, it turns into a chip conversation within 30 seconds. That makes sense. It is visible. It is easy to talk about. But it skips a step that is boring and expensive and, honestly, unavoidable.

AI has a physical bottleneck. Buildings. Power. Cooling. The stuff that keeps a data center alive at 2 a.m. when nobody is watching.

That is where Comfort Systems USA (NYSE: FIX) lives.

What’s interesting is how blunt the latest quarter was.

Q1 2026 revenue was $2.87B, up 56.5% from $1.83B a year earlier. Diluted EPS was $10.51 versus $4.75. Operating cash flow came in at $388.8M, a big swing from an $88.0M outflow in Q1 2025. And backlog was $12.45B as of March 31, 2026, up from $11.94B at December 31, 2025 and $6.89B at March 31, 2025.

I’m repeating the backlog line on purpose. It is the part people gloss over. A lot of companies can show a hot quarter. Backlog tells you whether the phone is still ringing.

Also, management called out 51% organic growth (same-store revenue). That matters. It is not just “we bought growth,” it is “customers are pushing work through the door.”

Here’s the thing. Comfort Systems is not a tech company. It is a contractor with real constraints.

Labor is the obvious one. Scheduling is a mess in fast-growing markets, and skilled trades do not appear overnight. If you have ever tried to get a solid HVAC crew to show up on a tight timeline, you already get the vibe. People talk about “moats” like they are magical. Sometimes the moat is simply: the good teams are already booked.

And data centers are not forgiving work. Redundancy, uptime, commissioning, change orders, timelines. It is not a place where a customer shrugs and says “close enough.”

At first glance, this sounds like a simple picks-and-shovels idea. And it kind of is.

But the nuance is that the “shovels” in this cycle are not cheap shovels. They are engineered systems, installed by scarce crews, in buildings where the customer is willing to pay for speed and reliability. That changes the economics.

Now the uncomfortable part. A stock can be a great business and still be a bad buy on a given day.

FIX has been rewarded for execution, and the market has a habit of over-rewarding anything connected to data centers. I’m not saying it has to fall. I’m saying you should assume expectations are high, because they are.

So if you are looking at FIX, I would keep it simple and slightly boring:

• If backlog keeps growing (or even stays elevated) while cash flow stays strong, the story is still intact.
• If backlog starts rolling over and management starts talking about delays, you will feel it fast.
• If labor tightness gets worse, margins can get squeezed even while revenue looks fine.

What I’m watching next is not a buzzword. It is the next backlog update and whether the sequential trend stays positive. Q1 ended at $12.45B. That is the anchor.

One last thought, and then I’ll get out of your inbox.

AI investing is getting weirdly abstract again. Models, tokens, platforms, roadmaps. Meanwhile, the physical world is sitting right there, taking purchase orders and filling schedules.

FIX is not the only way to play that. It is just one of the cleaner examples where the demand shows up in plain numbers.