Deep-pocketed tech companies have committed to spending more than $2 trillion on artificial intelligence buildout.

And many investors seem pretty sure there will be a big payoff as reflected in market valuations of “Magnificent 7” companies, including Alphabet (Google), Microsoft, Amazon, Meta (Facebook), Nvidia and Tesla leading the AI charge.

Valuations for nonpublicly traded OpenAI reached an estimated $500 billion based on a late 2025 share sale, with speculation pointing toward a $1 trillion initial public offering sometime soon.

Nvidia recently surged past $5 trillion and is now hovering around a mere $4.5 trillion — with Microsoft, Apple and Alphabet (Google) all joining the $3 trillion club.

Note that Apple was the first U.S. company to hit a $1 trillion valuation in August 2018, soon followed by Microsoft and Amazon. Check out price-to-earnings ratios that are creating these eye-popping valuations:

These valuations are believed to reflect their market dominance, significantly high growth expectations and strong recent growth. The Shiller P/E Ratio developed by Robert Shiller, designed to smooth out business cycle impact, shows a current P/E for the S&P 500 at 40.8 versus a long-term historical average of around 17.

That means valuations are more than double the historical average, not unlike peaks seen before both in 1929 and the 1999 tech bubble.

According to Shiller data, P/E levels eventually lead to market corrections, either through falling stock prices or rising earnings as seen in the past. Reversion to the mean.

The trillion-dollar question is which will occur — stock prices falling or earnings rising — to normalize price/earnings?

Research I’ve followed suggests that the answer to that question is whether AI becomes a general purpose technology (GPT) or not.

Vanguard’s 2026 Economic and Market Outlook provided some interesting insights into this by looking back at railroads in the 19th century, post-World War II industrial expansion and the internet/personal computers of the 1990s, all of which became fundamental general purpose technologies of our economy.

Vanguard’s study (based on Bureau of Economic Analysis data) showed that these previous capital buildouts occurred over multiple years — typically with 4-6 year windows — suggesting the AI buildout is still in early stage at perhaps 30%-40%.

At the center of this buildout is the race for computing power and data storage, with more than $2 trillion of additional capital investment.

It appears the AI scalers are good for it based on cash flows, but there are serious challenges in meeting the incredible power demands.

BloombergNEF forecasts U.S. data center power demands will more than double in a decade, and utilities are also predicting increased power demands from all other customers.

All this assumes the current state-of-the-art chip, but I continue to wonder if chip makers will develop new chips that use dramatically less power, which could dramatically change the picture.

All of this leads to some important questions: Will the tech companies overextend themselves with all this spending? Will the buildout frenzy result in overcapacity like we saw during the telecom boom in the late 1990s as companies raced to “corner the market” by laying fiber optic cables during the dot.com bubble only to find massive overcapacity, which was dubbed “dark fiber?”

While the fiber was eventually used, years of delay led companies like Global Crossing, Worldcom and others to file for bankruptcy protection.

Stories like this should not be ignored by investors; they sure are not being ignored by utility companies in negotiating with the AI companies, often demanding financial commitments to buy the new power whether it is needed or not.

The Vanguard report shows that AI startups outnumber public companies in the United States with 6,956 private AI startups compared to 4,010 publicly traded companies of all kinds.

History argues that early leaders rarely maintain their dominance indefinitely, and many don’t even survive.

Just think about hundreds of auto manufacturers that failed despite innovative ideas, including Packard, Duesenberg, Pope-Toledo, Rambler, Tucker and many more. Even longer lasting nameplates like Pontiac, Oldsmobile and Mercury have disappeared.

It seems almost certain that many of the nearly 7,000 AI startups will either be gobbled up or disappear, too.

At the heart of this story is a shifting narrative about broader adoption, integrating AI into existing cloud platforms and creating saleable products.

Will AI become a truly universal technology like railroads, the internet and personal computers? Will AI be widely adopted by businesses and lead to higher long-term productivity growth? Can AI dramatically reshape workflows, improve worker productivity and reallocate workers’ time to higher value tasks? Will all or just some of the AI tech companies produce a big payoff on the huge capital investment being made?

Vanguard lays out equity return prospects under three AI scenarios: 

1. AI’s transformation is stronger than anticipated (upside).

• 10% probability
• 8%-plus earnings growth
• 8%-10% 10-year annualized stock return projection

2. AI emerges as general purpose technology and generates a 3% trend in U.S. growth (Vanguard medium run baseline)

• 60% probability
• 6%-10% earnings growth
• 5%-7% 10-year annualized stock return projection

3. AI disappoints, and exuberance is irrational rather than justified (downside)

• -2% to 2% 10-year annualized stock return projections
• 30% probability
• 3%-5% earnings growth

Of course, there are many who are much more bullish about AI results and profitability than Vanguard. But I find their reasoning sound as the market may well be underpricing the potential for AI investment to underdeliver.

There is the potential for some AI to become a commodity and be priced accordingly.

The massive capital spending for scarce resources (chips and power) could take a toll on profitability. And current stock valuations may in fact be too high with a reversion to the mean taking a toll on returns.

All of this is a reminder about the importance of diversification when investing. Having bets of AI likely makes sense, but having some exposure to value stocks, developed non-U.S. stocks and high-quality bonds can provide valuable protection if the AI boom isn’t all it’s priced for.

Diversification can reduce risk, and may be valuable in getting a good night’s sleep.