Everyone’s obsessed with who wins the AI race. OpenAI versus Google versus Anthropic. It’s like watching a heavyweight boxing match where the stakes are trillion-dollar markets. But here’s the thing: you’re watching the wrong fight.
The real money isn’t in picking the knockout artist. It’s in owning the boxing ring.
Think back to the dot-com era. Pets.com crashed. AOL faded. But Cisco? Cisco made 3,400% in five years because every single byte of internet traffic needed its routers and switches. Cisco didn’t care who won the website wars—it got paid either way.
The same playbook is unfolding in AI right now, except the infrastructure bill just got way bigger.
Here’s why: AI is shifting from chatbots to agents. And that’s not a small upgrade—it’s a 20 to 30x multiplier on infrastructure demand.
A chatbot? You ask it a question, it spits out an answer. Done. Maybe a few hundred tokens processed.
An agent? That’s a worker. Tell it “grow our market share 15% this quarter,” and it goes to work. It researches competitors, pulls internal data, drafts campaigns, tests messages, coordinates with other agents, revises, and keeps iterating until the job’s done. We’re talking tens of thousands of tokens per task. More compute, more memory, more networking, more cooling, more power. Everything gets multiplied.
And this isn’t theoretical. Over half of major enterprises already have AI agents running in production. Adoption is about to explode.
So where does all that demand land? Think of it as six tollbooths every AI workload has to pass through:
Compute is the obvious one—the chips that actually run the models. Memory keeps the agent’s context alive as it works. Networking lets agents talk to databases, APIs, and each other at lightning speed. Thermal management handles the insane heat these systems generate. Power keeps them running 24/7. And real estate—all those servers need somewhere to live.
The numbers are already showing up in earnings. Google’s processing 16 billion tokens per minute, up 60% last quarter. Nvidia’s CEO says inference compute demand is already 100 times higher than expected. Hyperscalers are spending like crazy on infrastructure. Cooling backlogs are expanding. Power companies are signing long-term deals. Data center landlords can’t build fast enough.
The tollbooth companies aren’t hoping this demand shows up—they’re reporting it every quarter.
Here’s the kicker: whichever AI model wins, the infrastructure bill stays the same. Every model needs compute. Every agent needs memory. Every workflow needs networking. Every rack needs cooling. Every data center needs power.
That’s the invisible tax. And the companies collecting it get paid regardless of who crosses the finish line first.
Sure, there are risks. Some of these stocks trade at premium valuations. A capex slowdown would hurt. Some have customer concentration issues. But those are timing problems, not thesis-breakers.
The shift from chatbots to agents is real. The infrastructure consumption is multiplying. And a narrow set of companies collects revenue as that consumption rises.
Most investors are still trying to pick the AI winner. That’s the hard game. The easy game? Own the road everyone has to drive on.
