Remember when the AI arms race was all about who could buy the most Nvidia (NVDA) chips? Well, the battlefield is shifting. The next big fight isn't just about silicon; it's about something more fundamental: power. And according to one prominent investor, it's about to make Big Tech's spending plans look like they're from a different, much cheaper era.
Gene Munster, managing partner at Deepwater Asset Management, dropped a prediction this week that should make any investor in mega-cap tech sit up straight. He says next year's capital expenditure from the giants will be "meaningfully higher than where the Street is at." The catalyst? A seemingly dry policy announcement from Washington.
On Wednesday, the Trump administration unveiled a voluntary agreement—dubbed the Ratepayer Protection Pledge—with the who's who of technology. The list of signatories reads like a roll call of the AI era: Microsoft (MSFT), Amazon.com, Inc. (AMZN), Oracle Corp (ORCL), Alphabet Inc.'s (GOOG) Google, Meta Platforms, Inc. (META), OpenAI, and Elon Musk's xAI.
The deal has a simple, expensive premise: to stop AI data centers from driving up electricity bills for everyday Americans. To do that, the companies have to play by new rules. They must build, bring, or buy their own power. They have to foot the bill for any upgrades needed to the electrical grid to handle their massive appetites. Critically, they can't pass those costs on to regular utility customers.
Think of it as a "you break it, you buy it" policy for the national power grid. And Munster thinks this changes everything for their spending plans.
He laid out the numbers on social media. For 2026, he's forecasting that mega-cap tech capex will jump by roughly 65%. That's a staggering figure, though it's actually slightly less than the 70% growth seen last year. The real eye-opener is his 2027 outlook, where he parts ways dramatically with the Wall Street consensus.
"The Street is expecting mega cap capex to increase by 14%," Munster wrote. "I expect it to be 40%."
To put that in context, he noted the current growth expectations baked into some stocks: Alphabet (GOOG) at 13%, Amazon (AMZN) at 6%, Meta (META) at 13%, and Microsoft (MSFT) at 19%. His forecast suggests all those numbers are about to get a serious upward revision.
Munster frames this as a fundamental turning point. "Tech is happy to pay for power," he wrote. "This marks a shift in the AI Arms Race from silicon to power."
In other words, the bottleneck and the battleground are moving. It's no longer just about securing enough advanced chips; it's about securing enough electrons to make those chips do anything useful. This pledge, in Munster's view, will "increase our reliance on Google, Meta, Amazon and Microsoft"—he mentioned Meta twice in his post, perhaps for emphasis—and that "the cost will translate to higher capex for longer."
The backdrop makes his argument hard to ignore. The U.S. added a record 10 gigawatts of new data center capacity in 2025. To give you a sense of scale, one gigawatt can power about 750,000 homes. In December alone, the country saw the largest single-month increase in capacity ever recorded. All this building is driving electricity demand higher at a pace not seen in a generation. Total U.S. power consumption rose 2.8% year-over-year—the fastest growth rate in roughly 20 years.
So, the next time you hear about an AI breakthrough, don't just think about the clever algorithms or the powerful chips. Think about the massive, humming data centers they run in, and the vast amounts of power they consume. That's where the real money—tens of billions of dollars more than anyone expected—is now headed. The race for artificial intelligence has officially become a race for actual electricity.
