Are We in an AI Bubble?

Dear Reader,

Are we in an AI bubble?

That’s the question I get asked most often.

I was talking to a buddy of mine this weekend.

He’s a bond trader, or was one, and he still does a little trading for Wall Street.

He’s about ten years older than me, and he was around in the ’90s, same as I was.

He said it feels a lot like the dot-com bubble.

I told him it does.

But there are some very big differences between the AI bubble and the dot-com bubble.

And yes, we are in a bit of a bubble, especially in specific areas.

Building and Hoping vs. Building to Keep Up

When you think of a bubble, you think about building capacity and hoping the demand comes.

You imagine a future where it shows up.

With AI, you’re building capacity and hoping you can keep up with the demand.

That’s a very different dynamic.

Think about the telecom boom in the ’90s.

They laid fiber across oceans and along railroad tracks.

They used a lot of leverage to do it.

They believed the demand would come, and it didn’t.

The fiber went dark, and because those companies had over-levered themselves, they went bankrupt.

It turns out they were right about the need for all that fiber.

They were just too early by 25 years, and their balance sheets couldn’t wait.

Now here we are in 2026, and a lot of that fiber is being used, especially with all the video we stream.

What’s Different Today

Look at what’s happening now.

You’re Microsoft, Amazon, or Google.

Your clients are S&P 1000 companies.

They’re coming to you and saying, “We need capacity. Help us integrate AI into an organization with 100,000 people.”

So you’re racing to build the data centers and the tools, hoping to keep up.

Demand for model spend is up almost twentyfold.

Will that change?

Of course it will.

But at this stage, when I look at the facts, this doesn’t look like the 1990s.

Not yet.

Where It Does Look Like the 1990s

There are bubbles all over the place inside this build-out.

Semiconductor prices are skyrocketing in the US.

The semiconductor producer price index is up 26%, the fourth largest increase on record.

That index declined every year from 1994 to 2021, because of Moore’s Law.

Technology gets cheaper.

Remember when flat-screen TVs first came out?

You walked into your friend’s house and he had a $5,000 flat screen, and it was so cool.

Seven years later that set cost $1,000.

Now you can get one for $500 or $600.

That’s what tech deflation looks like, and it’s flipped.

Computers, peripherals, and smart home assistants are up 9.7% year over year in September.

Software and accessories are up 28%.

And memory is the epicenter.

We’ve talked about this.

Michael Burry has famously bet against Micron.

DRAM contract prices jumped as much as 98% in the first quarter.

Gartner projects roughly a 130% increase in DRAM and SSD costs by the end of the year.

I expect relief in late 2027.

That’s when a lot of Chinese and Korean capacity comes online and floods the market.

At that point it’s going to look a lot like the telecom build-out.

Those memory makers will build it hoping demand keeps up.

It will, to a degree, for a little while.

And then a lot of those Chinese memory makers are going to get blown up.

We’re also feeling the inflation downstream.

Memory prices for smartphones rose 300%.

The memory in a smartphone now costs more than the application processor, for the first time ever.

It’s even hitting autos, with higher costs for chips, screens, and safety features pushing up vehicle prices.

That’s the inverse of the tech deflation we saw from 1994 to 2021, when the price of everything kept going down.

So Are We in a Bubble?

In aggregate, I don’t see it yet.

When I read the quarterly reports, the hyperscalers are saying they’re trying to keep up with demand.

They’re spending about a trillion dollars this year because S&P 500 customers are coming to them in a stampede to AI up their businesses.

They aren’t building data centers and hoping people show up.

When they start building excess capacity, we’ll be in trouble.

I’m not seeing excess capacity yet.

Don’t worry, we’ll get to that phase.

For now, it’s like an ocean.

You see little bubbles popping up all along the surface.

We have those everywhere inside this giant build-out.

But in aggregate, like I’ve said a hundred times, the bubble doesn’t pop until the Mag Seven and the hyperscalers stop spending.

The hyperscalers are building because their customers are coming to them in a stampede.

So the question isn’t whether to bet on the bubble or against it.

It’s who gets paid while the build-out happens, whatever happens to any single AI stock.

And the first thing every one of those data centers needs is power.

The public grid is years behind.

So the big players are signing long-term deals for power built right next to their data centers.

Geothermal is one of the few sources that can do that around the clock with no fuel to buy.

One company has been building geothermal plants for 60 years, and Google signed a 15-year deal with it.

That’s the opposite of building and hoping.

Get the name and full breakdown of the company here>>>

Have a wonderful day.

I’ll see you tomorrow.

“The Buck Stops Here,”