Is the massive and growing investment in AI infrastructure worth it?

It's a loaded question that yields different answers, depending on who you ask.

One of the sloppier answers I often hear goes something like this: "Every quarter, mega-cap tech companies announce new huge investments in AI. But even after spending all that money, they've been reporting huge quarterly profits."

This view isn't technically wrong, but it obfuscates how companies account for the investments in their financial statements. While it is true that companies are spending massive amounts of cash on AI (as reported on cash flow statements), the bulk of these investments have yet to appear as accrued expenses on future income statements.

And it's the income statement that gives us the quarterly earnings we hear about in the news every three months. It's how we get a better sense of a company's ongoing profitability.

This is a wonkier topic than what I usually write about. But I suspect it'll become increasingly important to understand as these expenses become a bigger part of quarterly earnings announcements.

From capex to depreciation expense

For the past few years, mega-cap tech companies have committed increasing amounts of capital toward buying hardware and building the massive facilities needed to support booming demand for AI. These capital expenditures (capex) are already in the high nine figures and are expected to cross the $1 trillion mark soon. The size and speed of the investment are why these companies, including Amazon, Meta, Alphabet, Microsoft, and Oracle, are called hyperscalers.

The companies selling the products and services to support this buildout have seen an immediate earnings boom, which has helped bolster the stock market this year.

"The mega-cap U.S. hyperscalers are on track to spend $800 billion on capex this year, an increase of 94% vs. 2025," Goldman Sachs' Ben Snider wrote in a Sept. 18 note. "That spending is flowing through to the earnings of the AI infrastructure complex including semiconductor, tech hardware, industrials, and utilities companies, which are collectively accounting for roughly half of consensus S&P 500 earnings growth this year."

For the hyperscalers themselves, however, much of the recent attention has focused on the size of the investment and how these companies are financing it. And while mountains of cash have been flying out of these companies' windows, most of this investment has yet to show up as expenses on income statements.

Remember: Big investments don't hit income statements all at once during a single reporting period. Instead, companies depreciate them over the investment's estimated useful life, which usually lasts years.

In other words, the hundreds of billions of dollars spent on AI infrastructure will effectively be chopped up and spread out over years. And that means we'll see much more of it show up in the form of ballooning depreciation expenses in future quarterly earnings reports.

For investors, the size of the investments doesn't really matter. What matters is whether these companies will generate enough revenue and operate efficiently enough to stay profitable as they work off those depreciation expenses.

Don't expect a definitive answer today about whether this bet on AI will be profitable. It's hard enough to predict the next quarter. It's much more difficult to predict what'll happen over the next five to 10 years.

But the coming depreciation expense is at least a bit more certain because it is a function of measurable capex. And Goldman's Snider ran the numbers and cautions that the coming depreciation expense will be a major hurdle for earnings growth.

"Hyperscaler depreciation expenses will continue to increase as capex growth decelerates, further dampening the boost of AI investment spending to S&P 500 earnings growth," Snider wrote. "We estimate a drag from hyperscaler depreciation expenses on S&P 500 earnings growth of 5pp in 2027, offsetting nearly half of the 11pp boost to earnings from capex spending. By 2028, the drag from depreciation should offset the S&P 500 earnings uplift from continued capex spending." (Emphasis added.)

In other words, Snider expects the AI earnings tailwind to fade by 2028 as depreciation expense becomes a bigger headwind. (This is similar to what Morgan Stanley analysts cautioned in January.)

Whether these companies overcome this depreciation headwind will largely depend on the revenue they generate from the AI goods and services they sell. (The revenue matter is a whole other can of worms, which we'll have to discuss later.)

Zooming out

Depreciation and accrual accounting are pretty advanced topics, and I'm sure many investors will continue to struggle to wrap their heads around it all.

For what it's worth, this is basic stuff for top business executives and the analysts covering the companies. That is to say, it's not some secret that this earnings headwind is looming. In fact, analysts are already forecasting earnings growth to decelerate in 2027 and 2028, which in turn may help explain why stock market valuations have been cooling over the past year.

Some investors may be able to get away with not understanding how it all works. Under normal conditions, accounting ignorance can be bliss.

But if you're someone who watches a lot of financial news and reads a lot of financial commentary, it would suit you well to have a basic understanding of how depreciation works.

Because eventually, headlines about the AI capex earnings tailwind will fade, and we'll hear a lot more about the AI depreciation earnings headwind.