Wall Street Warning: Massive Depreciation Costs from AI Investments Could Hit U.S. Stock Earnings
Alina Collins
AI infrastructure's $1.3 trillion in capital spending is being deferred into multi-year depreciation, making current profit statements look exceptional; Morgan Stanley warns this 'golden window' will close, and a depreciation wave could systematically compress future earnings.
What does the "golden window" actually mean?
Chip suppliers book revenue the moment they ship. But buyers like Meta, Microsoft, and others spread that spending across years of depreciation. This means → the seller's profit is already on the books, while the buyer's cost has yet to show up on the income statement.
Morgan Stanley accounting analyst Todd Castagno calls the current phase a "golden window where everyone looks great": S&P 500 earnings growth is on track to top 20% for a second straight quarter, with semiconductors and AI infrastructure as key drivers.
In plain terms = the income statement right now is a "dessert-first" bill — the sweet part has been served, and the bitter part is still in the kitchen.
How big is the cost gap?
S&P 500 companies are set to spend roughly $1.3 trillion in capex in 2025. Meta, Microsoft, Alphabet, Amazon, and Oracle alone account for $412 billion.
These five hyperscalers' 2026 capex is estimated at about $760 billion, while their depreciation and amortization for the same period is only about $211 billion. This means → roughly $549 billion in costs have not yet hit the income statement.
In plain terms = the money is already spent, but the ledger will take years to "recognize" it. The wider the gap, the more room there is for future profits to erode.
Why is cash flow already sounding the alarm?
In 2026 the five hyperscalers' combined free cash flow is expected to plunge 91% to about $16 billion, while net income is projected to rise 25% to $506 billion.
Amazon and Oracle are forecast to post negative free cash flow this year; Meta barely stays above zero. This reflects an extreme divergence between reported profit and actual cash — a direct symptom of deferred depreciation.
Put simply = net income says "I'm fine," free cash flow says "I'm almost broke." The two numbers are telling very different stories.
Why can't even analysts get the depreciation figure right?
Most hyperscalers shifted from asset-light to capital-intensive models only recently, leaving limited historical data to work from.
Companies have wide discretion over asset useful lives and can change annual depreciation by adjusting those estimates. Many data-center builds also use off-balance-sheet financing — debt kept outside the reported financials — adding another layer of opacity.
Analysts' standard deviation on Meta's 2028 revenue forecast is just 4% of the mean, but for depreciation it reaches 24%. This means → the market agrees on "how much money comes in" but disagrees wildly on "how much depreciation goes out." David Zion, founder of Zion Research Group, warns that "consensus depreciation estimates may be systematically too low."
Is the market's bet on a "V-shaped rebound" sound?
Consensus expects capex growth to slow after next year, with strong revenue driving a V-shaped recovery in free cash flow — from about $16 billion in 2026 to $185 billion in 2028, then $387 billion in 2029.
Yet the S&P 500's forward P/E already sits around 22×, above the historical average — and that is before the bulk of depreciation hits earnings.
This reflects a market that has already priced in the good news but has not fully priced in the "late-arriving bill." Whether AI giants can generate revenue large enough to cover these costs is the central test of today's valuation logic.
Content is for reference only, not financial advice.