AI Isn't a Bubble. It's Infrastructure.

Hank Green posted a video titled “The State of the AI Industry is Freaking Me Out.”

In it, he walks through what he calls “the AI money machine”—a graphic showing how Nvidia essentially subsidizes the purchase of its own chips to inflate demand. He argues that the money being made from AI services is “an order of magnitude lower” than what’s being spent on hardware, creating what looks like a circular investment bubble that’s disconnected from real consumer demand.

He has a point.

The numbers are wild. OpenAI signs a $300 billion deal with Oracle for data centers. Oracle spends tens of billions buying Nvidia chips to build those centers. Nvidia invests venture capital in AI companies, who then use that money to buy… more Nvidia chips.

It’s circular. It’s massive. And it feels speculative.

But I think we’re looking at this through the wrong lens.

I’ve Been Saying This for Months

Back in June, I wrote a four-part series called Grid-Level Thinking where I made a simple argument:

AI isn’t magic. It’s infrastructure.

I called it “the grid”—raw electrical current that needs to be wired into your systems with intention, not just plugged into like a miracle appliance.

At the time, I was talking about individual workflows. How to think about AI at the ground level. How to redesign systems around a new utility instead of just bolting it on and hoping for productivity gains.

But here’s what I’m realizing now, watching Hank Green’s video:

What I was observing in my own work in June—AI as utility infrastructure—is exactly what’s playing out at the trillion-dollar industry level today.

The metaphor scales.

At the individual level, AI is the power grid. You’re the architect who needs to wire it in.
At the industry level, AI is a General-Purpose Technology. Companies are building the power plants and transmission lines.

Same pattern. Different scale.

So when people say “it’s a bubble,” I think they’re missing what I saw back in June: this isn’t a product. It’s foundational. It’s the new electricity. And the spending reflects that.

What If AI Isn’t a Product?

Most people look at AI—especially Large Language Models—as an end product. A chatbot. A service. Something you sell directly to consumers.

And when you look at it that way, the numbers don’t add up. Of course they don’t.

But what if AI isn’t a product at all? What if it’s more like… electricity?

Not as something you buy and use directly—but as a foundational input that powers the ability to produce other things.

In economics, this is called a General-Purpose Technology (GPT). And it completely reframes the bubble debate.

The Power Grid Wasn’t a Bubble

Think about the early 20th century build-out of the electrical grid.

Colossal upfront investment. Thousands of miles of copper wire. Power plants. Transformers. Massive infrastructure spending with no immediate consumer ROI.

If you looked at those numbers in 1910, you could’ve made a compelling case that electricity was a bubble. “Who needs that much power? We already have gas lamps!”

But electricity wasn’t about the lightbulb. It was about what came after the lightbulb.

Factories redesigned their entire layouts around electric motors.
Telecommunications became possible.
Consumer appliances created entirely new industries.
Computing became possible decades later.

The lightbulb was just the first obvious application. The real value was in the spillover innovations that electricity enabled across every industry.

AI is the same.

This Isn’t About Chatbots

The mistake people make is thinking AI is just for chatbots, content generation, and customer service bots. That’s like looking at the first power plant and saying, “How many lightbulbs do we really need?”

LLMs aren’t a single product. They’re a foundational layer that can be “plugged in” to:

Drug discovery — cutting years off R&D cycles
Legal analysis — reading thousands of contracts in seconds
Software development — making engineers 10-50% more productive
Medical diagnostics — analyzing scans faster than human radiologists
Supply chain optimization — managing complexity humans can’t track
Personalized education — adapting to every student’s learning style

Each of these applications creates real economic value. Not promised future value. Real, measurable ROI today.

JPMorgan Chase CEO Jamie Dimon said his bank is already getting $2 billion in direct benefits from AI. Not in five years. Now.

This isn’t speculation. It’s infrastructure spending.

It’s Not a Bubble. It’s a Build-Out.

When you view AI as infrastructure—as a general-purpose technology—the “bubble” narrative falls apart.

Yes, Nvidia’s valuation is massive. But it’s not based on hype. It’s based on the market’s belief that Nvidia is building the power plants and transmission lines for the 21st century’s most important resource.

The spending from Microsoft, Google, Amazon, and Oracle isn’t speculative. It’s a long-term capital expenditure to re-tool their entire infrastructure for a new era.

The spending looks circular because we’re in the infrastructure phase. Just like the electrical grid required massive upfront investment before consumers saw the benefit, AI requires a foundational build-out before the full economic impact becomes visible.

The Real Difference From the Dot-Com Bubble

Here’s the key distinction:

The dot-com bubble (1999) was built on companies like Pets.com—valued in the billions with no clear path to profitability, spending millions on Super Bowl ads just to get noticed.

The AI build-out (today) is built on companies seeing immediate, measurable productivity gains. Developers are 10-50% more productive with AI co-pilots. Legal teams are processing contracts in hours instead of days. Researchers are finding drug candidates faster than ever.

This isn’t a future promise. It’s present-day reality.

The “bubble” valuation is simply the market attempting to price in the combined value of all the future applications that this foundational technology will enable.

That’s not a bubble. That’s how general-purpose technologies are valued.

The Money Flow Makes Sense—If You Squint

Let’s revisit Hank Green’s concern about the circular money flow:

Nvidia invests in AI companies
AI companies buy Nvidia chips
Oracle builds data centers with Nvidia chips
OpenAI uses those data centers

Yes, it’s circular. But so was the electrical grid build-out. Power companies invested in electric appliance manufacturers. Manufacturers bought more power. Cities invested in grids. Businesses bought into electrification.

The difference is time horizon. The electrical grid took decades to show full ROI. We’re expecting AI to show consumer-scale returns in months.

That’s the disconnect.

What This Means for How We Think About AI

If you’re building with AI, investing in AI, or just trying to understand where this is all going—stop thinking of AI as a product.

Think of it as electricity for knowledge work.

It’s not about whether chatbots are worth $300 billion. It’s about whether a foundational technology that can be applied to drug discovery, logistics, law, medicine, education, and engineering—across every industry—is worth investing in.

The answer is yes.

And the “bubble” you’re seeing isn’t speculation. It’s the messy, expensive, circular process of building the infrastructure that will power the next century of economic growth.

From Individual Workflows to Industry Economics

Back in June, when I wrote about “wiring into the grid”, I was focused on helping individuals and teams think more clearly about how to use AI in their day-to-day work.

I argued that everyone has access now, but most people aren’t redesigning their systems to take advantage of it. They’re just “plugging in” instead of “wiring in.”

What I didn’t realize at the time was that this same pattern—treating AI like infrastructure instead of a product—would be the key to understanding the economics of the entire industry.

The trillion-dollar spending isn’t irrational. It’s what happens when you recognize that you’re building the foundation for the next era of productivity and growth.

As I wrote in June, “power without purpose is just a bill.” The companies spending billions on chips and data centers aren’t just burning money. They’re building systems designed to extract compounding value from this new utility.

We’re not in a bubble. We’re in a build-out.

And it’s just getting started.

This post extends ideas from my Grid-Level Thinking series, originally published in June 2025.