Morgan Stanley: Tech Giants' AI Capital Intensity Surpasses Dot-Com Bubble Peak, $520B Depreciation Pressure Looming
0xBroomberg
Morgan Stanley's latest report shows that AI capex-to-sales ratios at five hyperscalers have officially surpassed the dot-com bubble peak — with $520 billion in depreciation set to hit over three years, profit margins face a historic squeeze.
How extreme is the capital spending?
Amazon, Google, Meta, Microsoft, and Oracle are projected to hit capex-to-sales ratios of 36% in 2026, 44% in 2027, and 42% in 2028 — all above the 32% historical peak set by telecom during the dot-com bubble.
This means → the current AI investment cycle is burning cash at an intensity that has no historical precedent.
Factor in finance leases — long-term equipment rentals where the company bears most of the cost — and the numbers jump further: Microsoft's FY27 ratio rises from 50% to 64%; Oracle's from 115% to 189%.
In plain terms = Oracle is spending nearly $2 on AI infrastructure for every $1 of revenue — less an "investment" and more an all-in bet.
How fast is spending outrunning revenue?
Over the past nine months, consensus capex forecasts for the five hyperscalers in 2026–2027 were revised up by roughly $900 billion combined. By name: Google up 139%, Oracle up 175%, Meta and Amazon up 85% and 81%.
Revenue forecasts, however, have barely moved in comparison. This means → a widening "scissors gap" between spending and income — the money is going out, but the revenue hasn't caught up.
This reflects a core tension: the market believes AI demands massive investment, yet still lacks matching evidence of revenue growth.
How large are the off-balance-sheet obligations?
Purchase commitments — contracts signed but not yet delivered — across six giants total nearly $982 billion: Google leads at $332 billion, followed by Meta at $238 billion, Amazon $155 billion, Microsoft $142 billion, Nvidia $104 billion, and Oracle $11 billion.
Uncommenced lease obligations add another $822 billion. On top of that, operating leases already on the books total $179 billion, finance leases $86 billion, and unpaid capex buried in accounts payable runs roughly $110 billion.
In plain terms = the balance sheet is just the tip of the iceberg. These companies carry nearly $2 trillion in off-balance-sheet leverage — real operating risk far exceeds what the financial statements show.
What is the "depreciation tsunami"?
Morgan Stanley estimates that Microsoft, Oracle, Meta, and Google will rack up over $520 billion in cumulative depreciation charges from FY26 to FY28. Oracle faces the steepest climb — depreciation as a share of revenue rising from 7% in FY25 to 28% in FY28; Meta goes from 9% to 19%.
Much of the capex currently sits in "construction in progress" (CIP) — assets still being built and not yet in service — meaning depreciation hasn't started yet. This means → the real profit hit is being temporarily masked by accounting rules.
The evidence: CIP balances at Oracle, Meta, and Google grew roughly 200%, 90%, and 55% over the past year. Once these projects are completed and reclassified as fixed assets, depreciation charges will accelerate sharply.
What keeps margins from cracking?
Companies have only two paths to defend margins: either cut other expenses to make room, or grow revenue fast enough to offset the depreciation wave.
But the current scissors gap — capex forecasts running far ahead of revenue forecasts — makes the second path increasingly difficult.
This reflects the fundamental risk of the AI investment cycle: these giants are betting that future revenue will catch up with today's spending. If AI monetization falls short of expectations, the margin decline won't be gradual — it will be a cliff.
Content is for reference only, not financial advice.