Big Tech is in the middle of one of the largest capital spending binges in corporate history. Estimates suggest hyperscalers like Microsoft, Google, Amazon, and Meta could collectively spend over $600 billion on AI infrastructure in 2026 alone—data centers, GPUs, networking equipment, and power infrastructure.
The bulls call it visionary investment in the future. The skeptics are asking a different question: what happens to earnings when all that spending starts flowing through the income statement as depreciation?
Here's the issue. When companies buy equipment, they don't expense it all immediately. Instead, they depreciate it over its useful life—typically 3-5 years for servers and networking gear. So that $600 billion in spending won't hit this year's earnings fully. But in 2-3 years? It'll be a massive ongoing expense.
Some analysts are already warning that return on invested capital (ROIC) for these companies could decline as depreciation ramps up. If the AI revenue doesn't materialize fast enough to offset the costs, margins compress. And if margins compress while valuations stay elevated, multiples have to come down.
This isn't theoretical. We've seen this movie before.
During the late 1990s telecom boom, companies spent hundreds of billions building fiber-optic networks for the coming internet age. They were right about the trend—internet traffic did explode. But they were wrong about the timing and who would capture the value. Most of those companies went bankrupt. The infrastructure they built got bought for pennies on the dollar, and eventually other companies profited from it.
The dot-com bubble saw massive capital expenditures on servers and data centers that generated little near-term revenue. Again, the vision was correct—the internet did transform everything. But the investors who funded the initial buildout got wiped out. The second generation captured the returns.
This AI cycle has some important differences. The companies spending this money—Microsoft, Google, Amazon, Meta—aren't startups. They have enormous cash flows from existing businesses. They can afford to make losing bets without going bankrupt. And they're seeing real revenue from AI products already, unlike the telecom companies that built infrastructure for demand that didn't yet exist.
