Everyone is excited about the AI revolution. But one big question looms...
Can the companies funding the AI build-out turn all that spending into profits?
After all, the numbers are getting increasingly staggering...
Citigroup (C) recently upped its forecast for AI-related infrastructure spending. The financial-services giant now expects this figure to hit more than $2.8 trillion through 2029. That's up from a previous estimate of $2.3 trillion.
The budgets are so large that tech companies can't cover them with free cash flow anymore. They're issuing debt to finance all that spending.
In the first nine months of this year, tech companies raised $141 billion in debt. That's up from the $127 billion they borrowed in all of 2024.
Investors are questioning whether all this spending makes sense.
Keep in mind that companies can't spread these costs out over decades, like with a factory. Given how quickly chips and servers become obsolete, a three-year life is a fair assumption here.
That would mean about $1 trillion in incremental depreciation and amortization expenses over each of the next three years.
Then consider that the entire tech sector made about $783 billion in net income in the past 12 months.
So it's easy to see how the math starts to get ugly. The tech sector as a whole will struggle to make money unless it figures out how to generate revenue in completely new ways – and soon.
And there's another issue with spending...
It's called the "Red Queen problem."
In a part of the "Alice in Wonderland" story, Alice and the Red Queen run as fast as they can in a race. But the landscape moves with them. As the Queen explains to Alice: "It takes all the running you can do to keep in the same place."
And in business, this describes situations where companies often have to spend huge amounts of money just to stay competitive – not to get ahead.
With AI infrastructure, if one big company spends $50 billion, the others also have to spend $50 billion just to keep from falling behind.
Worse, the spending doesn't necessarily add to any one company's competitive moat. When they're all doing the same thing, no one gets ahead.
But the real issue here isn't just the spending...
Tech Needs a Clearer Path to Huge AI Monetization
In tech, companies make money at what's called the "application layer." That's an actual product or service that consumers or businesses will pay for.
So far, there's no clear path as to how application-layer AI revenues will cover all those infrastructure costs. We haven't found that "killer" application product yet.
In fact, as my colleague Ethan Goldman noted last month, a recent survey by MIT found that 95% of organizations were getting zero return on their AI investments.
And without a "killer" application product, the tech industry investing in AI is like building a five-star hotel before you have any guests.
I see two possible scenarios as to how this all turns out...
The first is that someone figures out how to make money from AI in a big way. But given the current levels of spending, that application needs to be huge – and it has to come fast.
As an example, Amazon (AMZN), Meta Platforms (META), Microsoft (MSFT), and Alphabet (GOOGL) expect to spend more than $360 billion on AI infrastructure this year.
Those companies have an average operating margin of around 30%. So that means they will have to generate something approaching a trillion-dollar business to pay for this.
To put that in perspective, total revenues for those four companies come in at roughly $1.5 trillion in the past 12 months.
This means they need 70% revenue growth that has to come out of nowhere. And again, it needs to come fast... since everyone is still spending.
The second scenario is an oversupply of AI infrastructure. In this scenario, companies' margins would collapse. And it all becomes one big, commoditized service.
Under that scenario, maybe one or two of the spenders would make money. But the real value would accrue to companies using AI, not those building it.
And we have some historical precedents of big infrastructure build-outs that resulted in low returns for the folks behind them...
A Look at Previous Infrastructure Build-outs
For example, on May 10, 1869, America completed the first transcontinental railroad. In a grand ceremony in Utah, workers marked the event by driving a "golden spike" into the rail line.
But just four years later, defaults on railroad bonds led to the Panic of 1873.
New rail lines had been built too fast. They outpaced demand. And returns on railroad investments collapsed.
Or consider the U.S. utility industry in the early 1900s...
In 1898, Samuel Insull of the Chicago Edison Company proposed a deal for the industry. Utilities would let states set their rates. In exchange, utilities would have monopoly control over their territories.
Because of this, utilities became a commoditized and low-margin business. But the trade-off worked.
As guaranteed monopolies, they could borrow money at cheap rates.
And by 1914, 43 states had utility commissions that regulated the power companies. This structure allowed the industry to expand across the country much faster than it could have under competition.
Then there's the telecom bubble in the late 1990s...
From 1996 to 2001, telecom companies spent almost half a trillion dollars to construct fiber networks across the country. That was about twice the amount the federal government had spent on the interstate highway system in previous decades.
The reckoning eventually came. By 2002, the global telecom industry had lost more than $2 trillion in value. According to some estimates, less than 5% of the fiber installed during the boom was ever used.
In each case, the infrastructure builders were the losers.
But the real winners were other companies that used that infrastructure to build entirely new businesses. And they managed to capture massive value.
Cheap transportation by rail and a national reach gave birth to catalog retailers Sears and Montgomery Ward. They created great wealth by connecting manufacturers to a national market of consumers.
And cheap, reliable power helped the Second Industrial Revolution take off in the U.S. With electrification, factories could redesign their entire production processes around efficient assembly lines.
Workers became far more productive. And for the first time, mass production was possible.
Ford Motor's (F) Highland Park plant was a prime example of this. Electrification enabled a moving assembly line. This cut the time to build a Model T from more than 12 hours to only about an hour and a half.
And cheap, abundant bandwidth led to several of today's biggest tech success stories...
Amazon grew from about $3 billion in revenue in 2000 to more than $670 billion today. Meanwhile, Alphabet grew from $80 million in revenue in 2000 to more than $370 billion today.
Meta didn't exist until 2004. But it now has nearly 3.5 billion daily users across its social media platforms. And it boasts nearly $180 billion in revenue.
Together, these three companies are worth more than $7 trillion. And all that value was built on infrastructure that the telecom companies lost money building.
Considering all this, I see two ways to invest in the AI build-out...
Finding the Ultimate Winners From AI
The first is in the tech companies themselves. And as always, I'll let the Power Gauge be my guide.
In terms of the "Magnificent Seven" mega-caps, the Power Gauge gives three of them a "bullish" or better rating right now. And the other four are in "neutral" territory.
And when it comes to the subsector exchange-traded funds...
The Power Gauge gives the SPDR NYSE Technology Fund (XNTK) a "very bullish" rating right now. And the SPDR S&P Software & Services Fund (XSW) currently earns a "bullish" rating.
Those are strong industry groups. So they're great places to look for individual stocks.
The second way to invest is to find companies using AI to solve real problems and improve their competitive positions.
This way, you're placing your bets on smart users... not big infrastructure builders. (In fact, this is the strategy behind my Breakthrough Investor publication. If you aren't already a subscriber, you can learn more about it here.)
The big question isn't whether AI will create value. It's who will capture that value.
When the dust settles, history suggests it won't be the infrastructure builders spending trillions today... It'll be the companies smart enough to use that infrastructure to solve real problems.
Good investing,
Joe Austin
