For decades, tech’s magic trick was simple: stay light, scale fast, and let someone else own the expensive stuff. No factories, no fleets, no physical footprint, just code, servers, and vibes. But the AI era is flipping that playbook on its head. Today’s biggest tech giants suddenly look a lot less like software companies… and a lot more like utilities.
The reason? AI wants everything, power, water, land, and every spare dollar that can be shoveled into data centers the size of small cities. The “software eats the world” prophecy didn’t end. It just got hungrier.
The New Capital-Intensity Kings
The most shocking entrants among America’s most capital-intensive businesses aren’t oil producers or power companies. They’re Oracle $ORCL ( ▲ 7.58% ) and Meta $META ( ▲ 0.32% ) . Wall Street expects both to pour extraordinary sums into chips, servers, and data center buildouts over the next year, with Oracle leading the pack thanks to its massive multiyear deal with OpenAI.
OpenAI itself would sit completely off the charts. Its trillion dollar infrastructure commitments make even today’s AI heavy hyperscalers look modest.
Meta, meanwhile, plans to shovel nearly half of its incoming revenue back into AI related infrastructure. Investors cheered news that Meta may cut some of its metaverse spending, not because they hated VR, but because freeing up dollars for AI capex is the new cheat code.
Even the real estate names on the list, Digital Realty and Equinix, are basically data center operators. That means every one of the most capital-intensive companies in the index is either a utility or tied directly to the AI data center boom.
Oracle’s Accounting Looks Wild, But It Isn’t Shady
Oracle is expected to show huge profits a couple years out while simultaneously burning through cash. That isn’t fraud, just the math of AI capex. When Meta buys millions worth of Nvidia chips, it records it as capex, and the impact drips into earnings through years of depreciation. Nvidia, however, recognizes the revenue instantly.
The bear case, voiced loudly by folks like Big Short legend Michael Burry, is that hyperscalers are overestimating how long these chips last, which could make depreciation hit harder in the future. But early data from analysts suggests GPUs can run profitably for about six years, which supports current accounting assumptions. Heat issues with Nvidia’s Blackwell chips remain a wildcard.
Either way, depreciation for America’s largest tech names is surging and will keep rising as the AI buildout expands.
The Asset-Light Model Has Officially Left the Chat
By 2028, Amazon’s depreciation as a share of revenue will be multiples higher than it was a decade ago. Microsoft, Meta, and Oracle show similar jumps. Apple and Tesla are notable exceptions. Apple is moving more slowly into AI, and Tesla’s fleet effectively outsources compute to customer owned vehicles, which keeps its reported capex intensity low.
Then there is Nvidia and Broadcom, the chip designers powering the boom. Thanks to past spending and soaring sales, their depreciation burden is actually shrinking relative to revenue. Ironically, the asset light tech giant of the future isn’t a cloud titan, it’s a semiconductor designer.
Bottom Line: AI Is Ending Tech’s Asset-Light Era
The new tech economy isn’t built on code alone, it’s built on concrete, copper, cooling towers, and silicon. The companies that once bragged about owning nothing are now pouring tens of billions into the physical world. And despite the costs, investors seem convinced. If AI is the next industrial revolution, then huge upfront utility scale spending might just be the price of admission.