AI Isn’t a Bubble. It’s Bubbly.

There's a growing class of AI commentary that sounds very sophisticated.

The hyperscalers are overspending. The financing is getting circular. OpenAI sits in the middle of a giant web of capital commitments. Data centers are expensive. Power is constrained. Valuations are stretched. Therefore: AI bubble.

I think that conclusion is increasingly coming from people who understand pieces of the technology extremely well while still not understanding what the technology is actually doing. And I don't mean that as an insult. I mean that because I just watched it happen to a guy who has spent more than two decades in serious enterprise technology.

The Andy Problem. A few days ago I wrote about my friend Andy Surujnarine. Andy is not some executive who discovered ChatGPT last Tuesday. He's spent 20-plus years in cloud, infrastructure, cybersecurity and enterprise technology. He's been a CTO. He's built and operated serious systems. He's used AI, vibe coded, experimented with models and was already planning to spend more time with agentic systems.

Then we sat down and actually started building. Within a few hours, his estimate of what was practical changed dramatically. We gave an agent responsibility for monitoring his inbox. We built a Windows application even though Andy isn't a Windows application developer. We took things that, intellectually, he already knew AI could probably do and compressed them from days or weeks into minutes.

Nothing magical happened. His understanding caught up with the capability.

That experience has been bothering me ever since, because I think it explains a surprising amount of the AI bubble conversation. Andy wasn't ignorant about AI. He was informed. He was technical. He was using it. And he was still underestimating it.

That's the problem. You can know an extraordinary amount about artificial intelligence and still have a badly outdated mental model of what it means in practice. I wrote at the time: You cannot be told what AI is. You have to use it.

And I increasingly think there's a dangerous Dunning-Kruger zone forming around AI where people know enough to confidently analyze the economics of the technology without having experienced enough of the capability curve to understand what those economics are attached to.

That's a problem when you're predicting the death of the industry.

AI Is Bubbly as Hell.

Now, let's be clear. There is froth everywhere. There are stupid valuations. There are stupid companies. There will be stupid data-center projects. There will be terrible loans. There will be stranded GPUs. There will be startups that raised $200 million to build something somebody can recreate six months later with three agents and a credit card.

Some of this stuff is absolutely going to blow up. Good. That's price discovery. But that does not mean AI itself is a bubble. It means we're going through an expansion so fast that capital, infrastructure and understanding cannot keep up with capability.

Look at the velocity. Every week we get new models, new agents, new chips, new tools, new modalities, new robotics systems, new ways of turning intent into execution. We don't even fully know what to do with everything we're producing yet. We're basically churning capability out at AI speed while the rest of civilization tries to catch up.

Of course it's bubbly. How could it not be?

Follow the Plumbing. The bigger mistake is treating all of this as though we're still debating whether people will use AI. We're long past that. The financial system is reorienting around it.

Wall Street has built infrastructure teams around data centers, power and AI. Private credit is moving in. Project finance is moving in. Utilities are planning around AI load. Governments are treating compute and power as strategic assets. Capital markets are developing ways to finance the physical infrastructure required to produce intelligence.

I've called that the financial plumbing of AI. And once the plumbing starts getting rebuilt, we're no longer talking about a novelty product. We're talking about infrastructure.

That's why the railroad analogy matters. Railroads were spectacularly bubbly. Companies failed. Investors got wiped out. Routes were overbuilt. Debt blew up. Financial panics followed. And afterward? We still had railroads. Because the failure of individual investments did not mean moving freight suddenly became unnecessary.

The dot-com bubble did the same thing. The bubble popped. The Internet didn't. Those are two very different events.

I Expect Bubbles to Pop. So I'm not predicting that nothing breaks. Quite the opposite. I expect lots of bubbles to pop inside AI. A startup valuation bubble. A GPU scarcity bubble. A data-center pricing bubble. Probably some ugly private-credit structures. Maybe even some companies we currently assume are untouchable.

Fine. Technology revolutions destroy capital all the time. What they don't necessarily destroy is the underlying capability. And AI has already crossed the line where the underlying capability is difficult to dismiss.

I've watched one highly experienced technologist recalibrate his understanding of what he can personally accomplish in a single evening. Multiply that across developers, lawyers, doctors, engineers, researchers, designers, entrepreneurs, analysts and every other person who hasn't had their own version of that evening yet. That's what the bubble analysis is trying to price. And I'm not convinced many of the people doing the analysis actually understand the size of that denominator.

Not a Bubble. Bubbly.

AI isn't a bubble. It's bubbly. It may be the most effervescent technological expansion we've ever experienced.

Capability is expanding faster than our ability to deploy it. Capital is moving faster than our ability to value it. Infrastructure is being built faster than our ability to forecast utilization. And people's mental models are aging faster than at any point I can remember.

That's going to produce wreckage. It should. But we're making a mistake if we look at the bubbles forming on the surface and conclude there's nothing underneath them.

The real risk in analyzing AI right now may not be believing too much. It may be confidently believing you understand the technology because you've been watching it closely. Andy was watching closely too. Then we built some shit. And that was different.

Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.

---

Sources

You Can’t Learn AI From the Bleachers — Rich WashburnOn why even experienced technologists can dramatically underestimate AI capability until they actually build with it. https://www.richwashburn.com/post/you-can-t-learn-ai-from-the-bleachers

Leopold Was Right: The Terrain Was Bigger Than the Map — Rich WashburnAI as a full industrial stack encompassing compute, power, infrastructure, capital and the physical systems beneath the models. https://www.richwashburn.com/post/leopold-was-right-the-terrain-was-bigger-than-the-map

Tracking the Stack: A Trail of Receipts — Rich WashburnTracks the movement of capital, infrastructure finance and major institutional players deeper into the AI buildout. https://www.richwashburn.com/post/tracking-the-stack-a-trail-of-receipts

The Factory Learned to Borrow — Rich WashburnLooks at the emergence of project finance and debt structures around AI infrastructure and the idea of financing long-term intelligence capacity. https://www.richwashburn.com/post/the-factory-learned-to-borrow

The Canary Is Dead — Rich WashburnOn the AI constraint shifting downward through the stack — from models and GPUs into power, infrastructure and financing. https://www.richwashburn.com/post/the-canary-is-dead

When Wall Street Builds a Team, Follow the Money — Rich WashburnExamines dedicated Wall Street teams forming around AI data centers, energy, cooling and infrastructure finance. https://www.richwashburn.com/post/when-wall-street-builds-a-team-follow-the-money-1

$710 Billion and Nowhere to Plug It In — Rich WashburnOn hyperscaler AI spending running into the physical limits of electrical generation, transmission and interconnection.https://www.richwashburn.com/post/710-billion-and-nowhere-to-plug-it-in

Financing AI Infrastructure and Data Centers — J.P. MorganOverview of the enormous capital requirements behind AI infrastructure and the mix of corporate debt, project finance and equity being used to fund it.https://www.jpmorgan.com/insights/banking/capital-markets/financing-ai-infrastructure-data-centers

Artificial Intelligence Investment Outlook — J.P. Morgan Asset ManagementInstitutional perspective on AI investment, infrastructure spending and the broader economic implications of the buildout.https://am.jpmorgan.com/us/en/asset-management/institutional/insights/market-themes/artificial-intelligence/

Data Center Electricity Demand to Grow 26% in 2026 — GartnerResearch highlighting the rapid growth in data-center electricity consumption and the increasing role of power availability as an AI constraint.https://www.gartner.com/en/newsroom/press-releases/2026-06-10-gartner-says-data-center-electricity-demand-to-grow-26-percent-in-2026

NVIDIA Earnings and AI Data Center Demand — Associated PressCoverage of NVIDIA’s latest earnings, continued data-center growth and ongoing supply constraints amid strong AI demand.https://apnews.com/article/dc8d556e709b50915cca9217a60b1991

NVIDIA Forecasts Revenue Above Estimates — ReutersReporting on NVIDIA’s latest quarterly outlook and continued demand for AI infrastructure and accelerated computing.https://www.reuters.com/business/media-telecom/nvidia-forecasts-quarterly-revenue-above-estimates-2026-08-26/