History is littered with technologies that changed the world while bankrupting those who paid to build them.

Today’s largest tech companies are spending unprecedented amounts on AI, and nobody knows if the returns will justify the investment.

We create portfolios around long-term cash flow. So I spend a lot of time thinking about the durability of business models, supply constraints, and capital expenditures (CapEx).

That is true whether we’re looking at real estate debt, private equity or dividend growth stocks.

Most investors overly focus on the income statement. Revenue growth is sexy. Few people give a flying hoot about the CapEx required to produce those gains.

But the balance sheet can hide large problems for a long time. A company’s CFO can:

So no cash flow, yet the income statement looks fantastic.

That’s how some companies can report strong revenue growth and rising earnings while burning mountains of cash to keep the machine running.

Side bar - apologies, but my son needs to hear this metaphor so bear with me. I imagine this is the equivalent of “instagram famous.” Looks impressive in filtered 2D but… not so much in the light of day. Exciting? Perhaps. But not a “long-term investment” Buddy (aka don’t you dare bring an “influencer” home to meet your mother).

Unfortunately, many businesses are like this. A lot of smoke and mirrors. They stay afloat and cover salaries but generate poor returns on invested capital (ROIC). They just spin their wheels.

Therefore, the key is to get your money in business models that defy financial gravity. Businesses with pricing power, limited competition, and long reinvestment runways. Aka unicorns.

High capital spend has to lead to high returns otherwise what is the point? If CapEx doesn’t slow relative to cash flow growth then you’re just subsidizing your customer (selling dollars for 80 cents).

This is why the Mag 7’s unprecedented CapEx push is so interesting to me. It will be the definitive case study for returns on invested capital. Either a brilliant gamble or a spectacular failure of capital allocation.

Most hyperscalers were capital-light businesses with insanely high returns on capital. Today they’re shoveling CapEx into a strategy without knowing what’s on the other side of this mountain of moolah.

I’m not sure most investors fully appreciate this new risk profile.

The hyperscalers are even starting to dabble with debt and issuing new (dilutive) equity to fund this spend. To be fair, most of the spend so far has come from cash flow.

Therefore, I’m not saying this is an insane bubble, or that they won’t continue to be great businesses. But it does appear their business models are slowly converging because of this arms race.

They have gone from monopolies in separate lanes to competing against each other in capital-intensive compute. That may be where most of the outsized profits from AI eventually accrue. This seems more likely than the AI model businesses (ex: OpenAI) which might end up building really cool commodities as the price of intelligence trends to zero.

As a long-term focused investor, I actually prefer hard asset businesses that have huge reinvestment runways. In other words pricing power and proven, profitable ways to spend CapEx which widens their moat.

The problem with AI so far, is we all know it’s amazing but we’re all just guessing on the payouts. The trillions in excess profits (on a timeline that makes the math worth it) remain hypothetical.

A high probability of return is the difference between investing and gambling.

That’s fine if you know the bet you’re making and size it appropriately (typically small).

It’s hard to know who’s going to win when market dynamics move fast. Look how quickly businesses are switching to Chinese open-source AI models to save on token costs.

Why are businesses going to keep paying premium prices for the latest frontier model tokens when the open source models are close enough?

AI models are starting to sound like a commodity where the smartest teams ever assembled are in direct and ruthless competition. That sounds like a market where the customer (us) is going to capture all the value.

History suggests new technology wins while competition gifts the economic value to customers through lower prices and higher productivity.

As an investor, that sounds less fun than owning a capital-light duopoly.

But if not the consumers, which companies will capture the long-term profits generated by AI across the value chain?

So far, it’s been the semiconductor companies.

Maybe this continues. However, those excess returns have brought out fierce competition as the hyperscalers start building their own chips.

This spending may or may not be irrational. These companies believe failing on AI is existential to their business; that they don’t have a choice.

The problem is the outcome has grown less certain while the expectations have grown more demanding.

And history has been unkind to CapEx investors in new technology. New tech often creates enormous value for society while lighting capital on fire for the investors who funded them.

Let’s revisit the graveyard of prior Tech CapEx cycles:

Railroads. Railroads transformed commerce and opened entire regions of the country. Aggressive construction, competition, and overcapacity produced a whole bunch of bankruptcies. Most of the benefit flowed to manufacturers, farmers, landowners, and consumers.

Telecom. Demand for communications exploded. Carriers spent heavily building networks and buying crazy expensive spectrum, then earned nonexistent returns as competition killed pricing and the debt just kept piling up on carrier balance sheets.

Fiber-optic cable. The internet buildout is the most obvious parallel to today. Everyone knew the world would never be the same. And we still blew it. We laid WAY TOO MUCH fiber than the market needed at that time, creating overcapacity and a price collapse.

Airlines. Commercial aviation changed the world and is responsible for unmeasurable gains in global productivity. And yet, what a staggeringly shitty business.

Insatiable need for capital, furious customers that expect flawless execution, no delays and low fares, excessive regulatory demands, and if you make a mistake, a lot of people die. No thank you.

Automobiles. Woof, see airlines. A CapEx consuming monster. Equally brutal business.

I’m sensing a theme. The common thread: People are heavily involved. Americans are unrivaled at exciting innovation and occasionally taking good things just a tad too far.

In each case the technology was transformational and demand was booming. Yet the investors who supplied the risk capital, who bravely brought these new technologies to life that benefited humanity, what was their reward? Most lost a metric ton of money.

This is a long-winded way of saying, being right about AI is not enough.

The counterargument is, “This time is different.”

Maybe. But I’d argue each of the above technologies was (on a relative basis) just as transformational at that time. It’s an impossible comparison, but to the world in 1800, the Railroads were pretty damn revolutionary.

The bullish case for AI will likely prove correct…eventually. It will raise productivity, reshape industries, and end up embedded in all business. None of this answers the question: is this bonkers amount of capital “Money Good” today?

Will the projected plus or minus $8 trillion (what’s a Trillion here or there between friends) earn attractive, durable returns?

Because recouping that capital is one thing, the bar for an acceptable return is much higher.

Even a relatively subdued 10% IRR on the AI investment requires more than 100% revenue growth per year over the next 5 years.

Of course, AI revenue could skyrocket and this will all look painfully obvious in the rearview mirror. But that’s not the point.

My beef is this is the largest bet made in the history of earth and the market has already decided the riches are just around the corner.

A rosy future is priced into most AI related stocks today. It’s consensus at this point. I agree the cash flow will come as well. But it’s the when and will it be enough that’s more in doubt.

The higher expectations climb, the less room you have for delays, pricing pressure, new competition and government regulations.

The major players in AI have proven to be exceptional investments. But will the current winners be as dominant the next 10 years?

AI does not need to fail for AI stocks to underperform. It only needs to take a bit longer and deliver a bit less value than share prices already assume.

High expectations are key to building a great life. But they are dangerous in investing. A sexy business plus sky-high expectations means high multiples on potentially peak earnings and no margin for error.

This isn't a case against owning any AI related stocks. It's a case against owning ONLY (or mostly) technology investments.

So we want a good chunk of money in the unsexy corners, where returns come from contractual cash flows and boring businesses quietly compounding.

AI is changing the world. That does not mean all of today’s investors will earn attractive returns funding the transformation.

Being right about the technology is only part of the equation. You also need to be right about who actually captures the profits, how much capital they must invest to maintain those profits, and how much optimism is already embedded in the price.

I'm not telling anyone to sell their tech. We own it too, and it's been a hell of a ride. This is a sizing question. It pays to know how much of your portfolio is riding on the same bet.

Brad Johnson

Evergreen Capital