"Where's the return on investment?"

That’s the #1 question investors are asking of the big artificial intelligence (AI) spenders right now.

Big tech companies are on track to spend roughly $750 billion this year on AI infrastructure… while they’re expected to rake in $175 billion in AI revenues.

That leaves a $575 billion gap.

• There’s one big problem with this comparison…

A large AI data center takes around two years to build and fill. The owner must secure land… get power… install cooling systems… and assemble thousands of servers.

Cash starts flowing out long before the first customer logs in.

So most of the AI revenue showing up today came from equipment bought in 2023 and 2024. Most of the $750 billion being spent this year won't earn a cent until 2028.

Comparing this year's spending to this year's revenue puts the cart before the horse. Companies like Alphabet (GOOGL) and Amazon (AMZN) also use their new chips to train models first. And training doesn’t directly earn them a cent.

Only later do those same chips switch to “inference,” a fancy term used to describe what happens when AI uses what it’s already learned to give you an answer.

That's when the cash register starts ringing.

• Does one chip pay for itself?

Nvidia's (NVDA) entry-level A100 server launched in 2020 at $199,000. Eight chips in a box. Run it flat out for five years, and each chip costs about 65 cents an hour—power included.

What does that same six-year-old chip rent for today? About $1.64 an hour. From Amazon, it’ll cost you $3.72 an hour.

In computing, old hardware usually turns into junk. But AI chips are aging like fine wine. Even Nvidia's four-year-old H100 chip has climbed. Rental prices bottomed last October, then jumped by around 40% in March.

• Amazon just announced its AWS cloud business grew by 37% last quarter…

Its fastest growth rate in more than four years.

Microsoft (MSFT) reported a 43% jump for its Azure cloud business. And Google's cloud revenue surged 82%.

These cloud platforms are nearly two decades old. Big businesses normally slow down as they grow. AI put fresh wind in their sails.

AI is also lifting businesses that already earn billions. Google's search revenue grew 17% last quarter. Meta Platform's (META) advertising revenue jumped 27%.

AI helps Google answer harder questions. It helps Meta show people ads they click. Those gains land inside existing revenue lines. So they never show up under a tidy heading marked "AI revenue."

The most interesting evidence comes from a century-old business truck freighter: CH Robinson Worldwide (CHRW) now produces freight quotes in 32 seconds. That used to take a person 1,545 minutes (about 25 hours).

Three-quarters of its smaller shipping orders are handled with no human touch at all. The company saves around 600 hours of labor a day.

Operating costs fell 12.6% in a year, which is HUGE for a low-margin business. And yet this doesn’t show up anywhere in AI “return on investment” calculations.

• If AI computing power was overbuilt, prices would be falling.

They're doing the exact opposite.

In January, Amazon raised the price of its AI cloud capacity by about 15%. That was the first increase in nearly 20 years of cloud prices only ever going down. It raised them again on July 1.

And listen to Oracle's Corp.’s (ORCL) Larry Ellison describe a dinner with Elon Musk and Nvidia's Jensen Huang:

I would describe the dinner as Oracle and me and Elon begging Jensen for GPUs. Please take our money. No, no, take more of it. You're not taking enough.

• Frontier AI labs were once seen as bottomless money pits.

But did you know Claude maker Anthropic is now cash flow positive?

It started this year at a $9 billion annual revenue run rate. By May, it had passed $47 billion—5X bigger in barely four months!

Anthropic also projects its first operating profit this quarter.

Demand is so high, it’s having to turn customers away. It capped usage and cut off partners because demand is eating computing power faster than it can buy it.

• There is one risk I'm watching…

The five biggest AI builders doubled their debt in five years, adding roughly $350 billion between them. Alphabet just posted its first negative cash flow quarter since it went public in 2004.

Too much borrowing is the real danger here.

But there's a big difference between now and the dot-com bust. Back then, telecom companies laid fiber-optic cables, and 85% of it sat unused for a decade.

Today, the equipment is sold out before the concrete sets.

• Follow the $750 billion…

While big tech giants are earning an ROI on their AI infrastructure, they’re still spending gobs of money.

That’s why, at RiskHedge, we prefer to own the companies drinking from this firehose of spending—not the ones doing the spending. In short, invest in the companies collecting checks from the buildout…

Nvidia sells the leading AI chips, and it’s trading at its lowest price-to-earnings ratio in years.

Taiwan Semiconductor (TSM) manufactures the most advanced chips.

Micron Technology (MU) supplies the high-bandwidth memory those chips need.

Plenty of smaller players are drinking from the same hose, including power producers… cooling specialists… electrical equipment makers… and data center component suppliers.

Stephen McBride

Chief Analyst, RiskHedge

PS: The AI buildout is creating a long list of winning stock investments, but there will be plenty of “hype” stories along the way that don’t hold up to scrutiny. In The Jolt, we can help you separate the winners from the losers. Go here to sign up if you’re not already a member.