Billions of dollars are pouring into the AI buildout, racing ahead of Wall Street forecasts and leaving analysts playing catchup.
A Barron's analysis of more than a decade of analyst forecasts shows capital expenditure estimates for hyperscalers including Alphabet, Amazon, Microsoft, Meta, and Oracle are being revised upward at historic rates.
Those miscalculations can cost investors. Last month, shares of Alphabet fell 7% the day after it announced plans to spend as much as $205 billion this year, up from its prior guidance of no more than $190 billion.
Other hyperscalers fell along with it, as investors grappled with how much Big Tech will have to spend to keep pace in the AI race. Hyperscalers collectively spent $330 billion in the first half of 2026, and analysts estimate that an additional $469 billion is being spent in the second half of the year.
The pace of spending is only expected to pick up next year, with analysts now estimating more than $1 trillion of capital expenditure across major data center operators. In reaching those figures, analysts have had to drastically revise their initial forecasts. Consensus capex estimates for the 2027 fiscal year have increased by an average of 183% across the five companies, compared with an average revision of just 33% over the past decade.
All five hyperscalers are experiencing their largest upward revisions in at least a decade. Alphabet's fiscal 2027 capex estimate has climbed from $82 billion to $301 billion-a 266% increase. Oracle's estimate has risen 262%, while Meta (162%), Amazon (147%), and Microsoft (78%) have all exceeded their prior highs.
Even with major revisions to their initial estimates, experts say that analysts are likely still underestimating future AI spending.
Beth Kindig, lead tech analyst at I/O Fund, wrote earlier this month that initial estimates for 2026 were "laughably low," and that projections for the coming year are "still too low." Goldman Sach's U.S. portfolio strategy team recently said that consensus capex projections for 2027 are "likely too conservative."
That spending has some investors nervous. In a note after Alphabet's latest earnings, Globlex Securities analyst Suwat Sinsadok wrote that "investors are being asked not to judge Alphabet's business, but to underwrite a spending program before its return is visible."
But Jessica Wachter, a finance professor at the University of Pennsylvania's Wharton School and a former SEC chief economist, says concerns over AI spending are likely overblown.
"We already know from the growth in the revenue of Anthropic and OpenAI that people are willing to pay for these services and... these model providers are the customers of these hyperscalers," Wachter says. "The dollars currently in the ground, building the data centers, I'm not sure how really risky that is. Of course, it's not without risk, but I kind of feel like it's more like the baseline level of risk that we all accept when we invest in the stock market."
Still, even if the revenue eventually justifies the current spending frenzy, history shows that early investors aren't always the ones who benefit from mass infrastructure buildouts.
That's the idea behind what John Turner, a financial history professor at Queen's University Belfast, calls the "useful bubble." The point is that while overspending on new technologies may not pan out for the investors, society as a whole can benefit. Turner says the rush of money associated with the dot-com bubble and with 19th century railroad investments led to the speedy buildout of important technologies and infrastructure.
In the meantime, companies that supply the hyperscalers stand to benefit from their massive infrastructure buildout. Beyond retail favorites like Nvidia and AMD, less familiar segments such as data center infrastructure and networking equipment generate a significant share of their revenue from hyperscaler spending.
In a recent report, Evercore ISI analysts identified CoreWeave, Western Digital, Seagate Technology, Arista Networks, and Vertiv Holdings as five companies with "high exposure" to AI capital expenditures. At Arista Networks, business with Microsoft and Meta alone represented roughly 42% of sales in the 2025 fiscal year, according to the report.
Since the start of the year, these five high exposure companies have had an average return of 100%, compared to the -3.6% average return from the five hyperscalers.
Experts say that dynamic is likely to continue over the coming years, as Big Tech continues to pour money into the AI buildout. And while hyperscalers may capture the headlines, the bigger investment story could be unfolding among the companies building the infrastructure behind what's estimated to be $1 trillion in annual spending-a figure, no doubt, that has room to grow.
Comments