A ticking time bomb awaits AI stocks in 2025.

• Barclays says companies face rising depreciation costs resulting from huge investments in integrated circuits.
• Barclays warns that these costs will have a material impact on profit estimates for leading technology companies.
• Depreciation costs could drive down AI stock prices and put valuations under scrutiny, according to Baird’s Ted Mortonson.

Companies capitalizing on the booming AI industry are racing to prove that their massive GPU investments are paying off, but there’s a rarely discussed problem that’s going to make that even harder.

Barclays analysts in a recent note said the depreciation associated with massive investments in AI chips is a “not-so-hidden” cost of AI that few investors factor in when analyzing the valuations of these companies.

Depreciation is an accounting method that lets companies spread the cost of a capital investment over its useful life. That means when a large-cap tech company buys billions of dollars worth of GPU chips, it doesn’t record it as an expense right away, but rather as a capital expenditure.

This can generate large profits right from the start, as capital expenditure is not immediately reflected in the company’s profit and loss account, but is recorded as a depreciation expense over the useful life of the asset.

The problem is that the useful life of AI GPUs may be much shorter than many expect, especially as AI chips go through an increasingly rapid innovation cycle, leading to higher-than-expected amortization costs that ultimately erode profits.

GPU amortization costs will be so high that Barclays is cutting its earnings estimates for cloud providers Alphabet, Amazon and Meta Platforms by as much as 10% early next year.

“The amortization of AI compute assets is the biggest expense for these leading companies,” said Barclays internet analyst Ross Sandler. “We think this is a risk that could rear its ugly head as we start looking ahead to 2025, so we’re flagging it early.”

As big tech companies spend hundreds of billions of dollars on expensive GPUs like Nvidia, amortization costs will increase over the next few years, especially as Nvidia moves to a one-new-product-per-year launch rate.

“Because Nvidia has a very aggressive design cycle of about a year between releases, all of these products have different feature and power profiles,” Baird managing director and technology strategist Ted Mortonson told Business Insider.

“It’s a headwind,” Morton said, adding that it’s big enough to impact valuations and cause AI stock prices to fall over the next year.

Barclays estimates that prevailing opinion on Wall Street is underestimating the scale of depreciation costs over the next two years.

For example, the bank expects Alphabet to record $28 billion in depreciation and amortization charges in 2026, up 24% from the current consensus estimate of $22.6 billion.

For Meta Platforms, the discrepancy between Barclays and Wall Street amortization estimates is even wider at $30.8 billion and $21.0 billion, respectively, meaning potential costs will be 47% higher than expected in 2026.

“In our view, given the modeling flaws, GOOGL, META and AMZN stocks are between 5% and 25% higher than consensus estimates,” Barclays’ Sandler said.

He added: “While we don’t believe valuations are stretched compared to historical bubble eras like 2021, the AI ​​boom has shed a clearer light on whether multiple expansions by big tech are justified, so it’s against this backdrop that depreciation (and therefore valuation) discrepancies are likely to come under scrutiny.”

One accounting method that CFOs of large technology companies are using is to extend the useful life of server assets from five years to six years or more, which helps spread costs over a longer period and mitigate the decline in profits.

But even that has its limitations, given the pace at which Nvidia releases new GPUs.

“We don’t see any megacaps extending the usable life of servers beyond this 6-year timeline because GPU cycles are getting longer quickly. As a result, megacaps will likely have to absorb higher depreciation costs going forward, unlike the last few years when we’ve seen adjustments to the usable life,” Sandler explained.

For Mortonson, it all comes down to the return on capital invested in AI.

“Wall Street has a big question. They’re spending over $200 billion now, and their CAPEX is up over 50%. Where’s the return on invested capital?” Mortonson asked. “We’re so early in this that, combined with all the accounting, it all comes down to return on invested capital, and I don’t think you’ll see return on invested capital until 2025 or 2026.”

Mortonson added: “I think the jury is still out. I think the accountants need to look into this, and there needs to be a lot more clarity between extending the lifespan of networks, storage and servers versus GPUs. That’s the heart of the matter.”