Will AI Blast Off or Blow Up? These Three Things Will Decide

In partnership with

[Note From Bill:  In yesterday’s kickoff to the new “What Comes Next” series, I looked at the Bandwidth Bubble of the late 1990s – as a precursor to the hyperscaler building boom. Today, in Part II, I look at the costs and risks of the $7 trillion Great Hyperscaler Race.]

I was looking for a flick to keep me company during a late work night last week and came across an old Warner Bros. favorite called The Great Race, a 1965 comedy that was nominated for five Academy Awards (winning one). A slapstick take on the famous 1908 New York-to-Paris Auto Race that supercharged global interest in cars, The Great Race made some splashes of its own, thanks to:

• An all-star cast (including Jack Lemmon, Tony Curtis, Natalie Wood and Peter Falk).

• True mechanized madness — with cars, airplanes, airships, submarines, racing boats and steam locomotives from that bygone era.

• One of the coolest names ever (Professor Fate) for a villain.

• A scene that movie buffs still refer to as “the greatest pie fight ever.”

• An ending (spoiler alert!) where an errant cannon shot brings down the Eiffel Tower (no snarky comments, please).

• And a $12 million price tag that (at the time) made it the single most expensive comedy ever filmed. (That’s a big deal: We’re talking about $125 million in current-day dollars — or what it cost to produce 2000’s Mission Impossible II.)

I laughed at the timing — since I was researching another “Great Race” to talk about with you.

And it’s another super-expensive race, at that.

But this one’s no comedy.

I’m talking about The Great Hyperscaler Race — the global buildout of data centers, power plants and the power-and-data networks that experts believe we’ll need to keep the Artificial Intelligence (AI) Era from stalling — or flaming out altogether.

White-shoe consultant McKinsey & Co. says companies will have to invest nearly $7 trillion in data center infrastructure between now and 2030 to meet the surging demand for AI — a technology that’s stampeded into virtually every corner of our modern economy.

To break that down a bit: We’re talking about more than $5.2 trillion for AI-related data center capacity and $1.5 trillion or more to piggyback existing IT workloads.

We’re talking about a lot of cash. A lot of opportunity. And a lot of risk.

Here’s some crucial context.

The Biggest Price Tag Ever?

Let me say this right here — and right now. I think the AI Era is the greatest tech story of our lifetime.

And as a four-decade veteran of finance, investing and journalism, I have the perspective to say so.

The tech boom I’ve lived through, worked through, written about and analyzed is like one of those giant cakes that get baked for a big wedding, an inauguration or a corporate launch: They’re baked and assembled in layers — and the final creation is something stunning to behold.

That “layered cake” is just what I’ve seen with the technology revolution that started back in the early 1980s.

I earned my journalism degree from Penn State writing stories on mechanical and electric typewriters. I started my newspaper career in 1984 just as the PC Revolution got started.

A decade later, the Internet boom came along. That was the “layer” that truly unlocked the power of the PC: We had to string all those PCs together — so we could talk with each other, work together and share our lives in real time — for the personal computer’s true magic to be seen.

The Cloud was the next “layer.” As well as all those connected devices.

AI is the latest layer. With quantum computing still to come.

Bring it all together and you’re talking about a towering layer cake more enthralling (and more expensive) than anything you’ll find on the Food Network. (Bobby Flay could never beat this one.)

So, yes, I’m long-term bullish on AI and quantum computing to come. But I’m also a realist, a Contrarian Investor and an objective analyst.

Wealth Killers see only one side: They’re bullish or they’re bearish — and can never seem to see the “other side.”

Wealth Builders see the whole playing field. They play the long game. But they take time to see what’s right in front of them.

And right now, we see some risk.

That projected $7 trillion outlay is a staggering sum. You could combine the economies of Japan ($4.28 trillion) and Canada ($2.23 trillion) — and still come up short. It’s equal to each of the market values of the energy ($6.75 trillion) and consumer staples ($6.11 trillion) sectors.

And it dwarfs the market caps of some key drivers (and beneficiaries-to-be) of The Great Hyperscaler Race — companies that offer contract web services, including:

• Nvidia Corp. $NVDA ( ▲ 1.77% ): $4.7 trillion.

• Apple Inc. $AAPL ( ▼ 0.2% ): $4 trillion.

• Microsoft Corp. $MSFT ( ▲ 1.37% ): $3.8 trillion.

• Alphabet Corp. $GOOGL ( ▼ 0.78% ): $3.5 trillion.

• Amazon.com $AMZN ( ▼ 1.22% ): $2.6 trillion.

• Broadcom Inc. $AVGO ( ▲ 0.73% ): $1.7 trillion.

• Meta Platforms $META ( ▼ 0.07% ): $1.5 trillion.

• Oracle Corp. $ORCL ( ▲ 2.43% ): $650 billion.

• And Alibaba Group Holding $BABA ( ▼ 3.78% ): $390 billion.

You get the idea.

We’re talking about an insane amount of money.

The Great Race was the most expensive comedy ever filmed back in 1965. Sixty years later, McKinsey says The Great Hyperscaler Race is “one of the largest technology infrastructure build-outs in history.”

That’s an understatement. It may be the most-expensive project — of any type — in modern history.

Even adjusted for inflation, nothing comes close:

Heck, even the Marshall Plan — the initiative to rebuild all of Europe after it was flattened by World War II — wasn’t remotely this costly. The United States spent $13.3 billion in 1948 dollars — equal to $179 billion today. Viewed another way, that $13.3 billion was equal to 5% of America’s then-GDP of $258 billion — which would be $1.6 billion of today’s $31 trillion U.S. economy.

The only thing I could find that even approaches The Great Hyperscaler Race was in the oil-and-gas sector, where capital expenditures are projected to grow from about $654 billion this year to $799 billion in 2030, says Mordor Intelligence. That gets you to roughly $4.3 trillion. Add in midstream and downstream investments, and we max out at $5 trillion to $6 trillion.

But once you push a bit, even this fails to come close. Some of that money goes to new exploration, but quite a bit is maintenance – to maintain supply and meet demand.

The Great Hyperscaler Race is a wholly new project.

That means the costs are high.

And so are the risks.

Risk No. 1: Overinvesting

Over-investing (spending too much too fast) is a major land mine and is the definition of a “bubble.” We’re talking here about ramped-up outlays in land, chips, data centers, power plants, networks, cooling systems, workstations — only to have demand fall short of all this processing “supply.”

Do that and you end up with something called “stranded assets” — like throwing a party but having nobody come.

As I showed you folks yesterday, this is just what we saw in the late 1990s, back during the Internet Frenzy, when massive outlays in high-speed, fiber-optic networks for which there was insufficient demand – creating a “Bandwidth Bubble” that collapsed in spectacular fashion.

It took a good decade for the massive overhang of bandwidth capacity to be absorbed — though (as we told you) it served as an early table-setter for the Cloud Computing Boom of the 2010s.

Telecoms spent an estimated $500 billion back then — just under $1 trillion in current-day dollars — to inflate that bubble. So it doesn’t come close to The Great Hyperscaler Race. And, yet, look at the ruinous financial fallout, with ripple effects that lasted a decade. The bursting of that bubble helped tip the similarly overinflated stock market into a collapse that caused a 78% wipeout in the Nasdaq Composite Index.

From a peak of 5,048.62 on March 10, 2000 – during the Dot-Com Bubble — the Nasdaq plunged all the way down to 1,114 in October 2002.

It took 15 years — until March 2, 2015 — for that tech bellwether to match its Dot-Com high. That, too, is the definition of a bubble.

While there are similarities between then and now, there are a lot of differences, too. A lot of the Internet companies were ideas in a corporate shell — no profits, no revenue and not even a real business plan. So there were a lot more speculative excesses that drove that Bandwidth Bubble.

But we’re talking a much bigger dollar number this time around.

And that’s where consultant Bain & Co. sees the biggest worries.

Bain says hyperscalers and chipmakers will drive AI’s next wave. But it says the journey will be a volatile one, that faces execution challenges and that “monetization” — getting a payoff that represents a return on the outlays — is the biggest risk.

Bain warns of an $800 billion revenue shortfall risk by 2030 if companies are getting less back than they’re pouring into AI infrastructure. With how AI demand is tracking, Bain says we’ll need to invest $500 billion a year to make sure supply and demand are in equilibrium by the decade’s end.

To sustain that — to make it work — it says companies will need to generate $2 trillion a year in revenue.

That’s something to watch — thanks to Risk No. 2.

Risk No. 2: The Revenue Shortfall

In its GenAI Divide Report for 2025, Massachusetts Institute of Technology (MIT) researchers found that lots of companies have yet to see an “AI payoff.”

U.S. companies have poured $30 billion to $40 billion into AI initiatives, but 95% report a zero return on investment (ROI) on those outlays. Only 5% — a microscopic percentage — say they’ve reaped a significant value, often millions in extra revenue or slashed costs.

The reasons are legion — and warrant a story of their own.

An assessment by Entrepreneur magazine said those winners tend to focus on one pain point, are well executed and are often done in partnership with external vendors — and not built wholly in-house.

We need a little bit of nuance here.

Speaking of “generative AI” — the “reactive” models used to create content — researcher Gartner says expectations were too high, ROI is elusive and lots of pilot projects fail.

“Generative AI is sliding through the trough of disillusionment due to a mismatch between high expectations vs. reality,” Gartner Vice President Arun Chandrasekaran said last year.

But “AI Agents” — proactive automation that’s the next phase, something I wrote about in my MBA thesis 30 years ago — will pack a payoff punch. Consider AI Agents another of those cake “layers” that I talked about.

In short, the story isn’t that AI is a “failed technology.” Companies are still figuring out how to best use it.

Even Gartner says the spending wave will continue.

It says AI investments will hit $1.5 trillion this year and blow past the $2 trillion mark in 2026.

In fact, it expects IT spending to exceed $8 trillion by 2030 — noting that AI infrastructure will be the major “Trigger,” though it doesn’t provide a “breakout” number.

That brings me to the third and final risk (as always, I’m intentionally keeping this super simple).

Risk No. 3: AI Unplugged

In the Bandwidth Bubble of the late 1990s, telecom companies built thousands of miles of high-speed fiber-optic networks — only to have the expected demand never develop (or, at least, not develop for another 10 years).

With The Great Hyperscaler Race, the biggest fear isn’t overcapacity that leads to those “stranded assets.” The worry is that all these data centers will get built — but owners will have no way to “plug them in.”

In short, industry execs fear that power will be a bottleneck. Before they can install all the computers to run AI programs, companies need what’s known as a “warm shell” — essentially a campus that already has power, cooling and the connections to the outside world.

And there aren’t enough to go around, says Satya Nadella, CEO of Microsoft.

You actually may have “a bunch of chips sitting in inventory that I can’t plug in,” Nadella said in a recent interview. “In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.”

Some hyperscaler projects being built now will consume 20 times the power of their already operating predecessors. Some individual “campuses” are projected to need as much as 2 gigawatts — the total power demand of some U.S. states.

According to one report, U.S. data centers are already consuming 4% of U.S. electricity. That’ll double by decade’s end, when AI-specific tasks could demand as much energy as 22% of American households.

This artist’s rendering of a data center project up near Wilkes Barre, Penna., shows where this will probably have to go. Yes, those are cooling towers for the site’s own nuclear power plant. That’s the Susquehanna Steam Electric Station, which Amazon Web Services wants to use to build out this data center.

Consultant Deloitte estimates U.S. AI data center power demand could grow 30 times by 2035, from 4 GW to 123 GW. Some planned campuses could require 5 GW each, more than the output of the largest nuclear plants.

Permitting is a byzantine process – with waits for interconnections in some areas taking as long as seven years.

Then there are the climate and water effects – stories unto themselves.

But power is right now the “defining constraint” of The Great Hyperscaler Race.

And it could lead to a de facto “underinvestment” in AI assets — the opposing extreme to Risk No. 1, and the exact opposite of the Bandwidth Bubble of the Internet frenzy.

The Bottom Line: As a Wealth Builder, I’m bullish on AI. I see the massive opportunity here.

But Wealth Builders are also objective. We’re candid about the risks. After all, forewarned is forearmed.

In Part III of my “What Comes Next” look at the stock market, the economy, big storylines and the opportunities and risks to watch for 2026, I’ll share some strategies and specific companies to look at for long-term Wealth Builders like you and me.

Wealth Builders run a great race. And they win. That’s what we’re here to help you do.

See you next time;