Meta's Free Cash Flow Just Fell 91%. The Real Story Is an Accounting Trick

Meta posted $60.8 billion in quarterly revenue, up 28% year-over-year. Its stock should have popped. Instead, investors zeroed in on a single number buried in the cash flow statement: free cash flow of $784 million, down from $8.5 billion a year ago. That's a 91% collapse in the middle of the best revenue growth Meta has posted in years. If you've been following the AI capex story across Microsoft, Meta, and the rest of Big Tech, you already know spending is up. What's less understood is the accounting mechanism sitting underneath these numbers — and it's worth understanding before you decide whether this quarter's stock reaction was an overreaction or a warning sign.

Where the $8 billion actually went

Meta spent $31.1 billion on AI infrastructure capex in the quarter and absorbed another $2.4 billion in legal charges. That's what turned a revenue beat into a cash flow bust. For full-year 2026, Meta guided capex to $130-145 billion, narrowing the range upward from the prior $125-145 billion floor — a signal management is doubling down, not backing off. Here's the thing that gets lost in the "AI spending is scary" headlines: capex hits the cash flow statement the moment the check is written. Depreciation — the accounting charge that actually dents reported earnings — gets spread out over years. That gap between cash going out the door now and the expense showing up on the income statement later is the entire ballgame for how these stocks get valued.

The useful-life trick nobody outside 10-K filings talks about

In January 2026, Meta quietly extended the useful life of its servers and networking equipment from four years to five and a half years. That single accounting change cut 2025 depreciation expense by $2.9 billion — money that never left the company, but also never showed up as a cost against earnings. This isn't a Meta-only story. It's become a pattern across the hyperscalers, and it's exactly the mechanism Michael Burry has been pointing to in his public critique of AI accounting: if you stretch out how long you claim a GPU or server rack will last, you shrink the annual depreciation bite and inflate near-term earnings, even though the hardware itself doesn't actually last longer just because the spreadsheet says so. GPUs in particular are widely understood to have shorter functional lifespans than general-purpose servers because each new Nvidia generation makes the old ones less cost-competitive to run, not because they physically break.

Photo by Pexels on Pixabay

Photo by panumas nikhomkhai on Pexels

Why this matters more than the headline capex number

Across the five largest hyperscalers, combined 2026 capex is tracking toward roughly $725-760 billion, while the depreciation and amortization they're actually recognizing this year is closer to $211 billion. The other half-trillion-plus sits on the balance sheet, waiting to become a real expense once assets go into service. That's not automatically bad — it's how capital-intensive buildouts have always worked, whether it was telecom fiber in the 1990s or oil refineries before that. But it means today's reported earnings for Meta, Microsoft, and Alphabet are flattered by a depreciation schedule that assumes GPUs stay economically useful for 5+ years. If that assumption is wrong — and Nvidia's roughly annual generational upgrade cycle argues it might be — the depreciation catch-up hits earnings later, all at once, right when the market has gotten comfortable with today's margins. As I covered in Microsoft and Meta Both Bet Big on AI. Wall Street Picked a Winner, the market has already started differentiating between the hyperscalers on execution. This depreciation mechanic is one more lens for that differentiation — it's not just about who's spending the most, it's about whose earnings are more dependent on favorable accounting assumptions holding up.

What would actually change my mind

I'm not calling this a fraud or a ticking time bomb — extending useful life estimates is legal, disclosed, and sometimes genuinely justified by better cooling and maintenance practices that do extend hardware life. What would make me more comfortable:

• Meta's AI-driven ad-targeting and Reels monetization numbers keep growing fast enough to justify the spend regardless of the accounting
• Capex growth decelerates in 2027 rather than compounding again off this year's already-elevated base
• Management stops extending useful-life assumptions further — a third extension in as many years would be a real red flag

What would worry me more: if depreciation extensions become an annual ritual across the hyperscaler group every earnings cycle, that's a tell that reported margins are increasingly doing accounting work rather than reflecting the underlying business.

The trade, if there is one

Meta's 10% post-earnings drop already priced in some of this skepticism, which means the stock is arguably a cleaner read on ad-business fundamentals than it was a week ago — the market is watching the capex/depreciation gap closely now, not ignoring it. If you're long Meta, the thing to actually track isn't next quarter's capex guidance, it's whether useful-life assumptions get stretched again and whether Reels/AI ad tools keep translating into real revenue growth rather than just narrative. This is not financial advice — always do your own research before making investment decisions. If you want one number to watch next quarter, it's not capex — it's whether Meta's depreciation expense as a percentage of revenue starts creeping back up. That's the moment the accounting cushion runs out.