Morgan Stanley says AI capex intensity now exceeds dot-com bubble peak

Morgan Stanley Warns AI Capital Spending Has Surpassed Dot-Com Bubble Intensity

The artificial intelligence infrastructure boom has entered uncharted territory. Morgan Stanley issued a fresh warning on Tuesday that AI-related capital expenditure among the largest hyperscalers has surpassed intensity levels seen at the peak of the dot-com bubble, raising questions about earnings sustainability as a massive depreciation wave looms.

Spending Beyond the Dot-Com Peak

The report, authored by analyst Todd Castagno, found that the cash capex-to-sales ratios for the five largest AI spenders — Amazon, Alphabet, Meta, Microsoft, and Oracle — are projected to reach 36% in 2026, 44% in 2027, and 45% in 2028 when accounting for lease obligations. Those figures exceed the approximately 32% capex-to-sales ratio recorded during the dot-com era’s peak.yahoo

Morgan Stanley has repeatedly raised its spending forecasts for the group, now expecting roughly $805 billion in combined capex this year and $1.12 trillion in 2027 — more than four times what was spent in 2024. Spending revisions for 2026-27 have risen by $630 billion compared to six months prior, a pace the bank called “unprecedented”.linkedin

The Depreciation Reckoning

Beyond the headline spending figures, Morgan Stanley flagged a structural risk that consensus forecasts may be underappreciating. According to a report first detailed by the Financial Times, the bank projects that Microsoft, Oracle, Meta, and Alphabet could collectively book more than $680 billion in depreciation charges in the coming years.ft

Oracle faces the most acute pressure. The company’s depreciation expense is expected to surge from 7% of revenue in fiscal year 2025 to 28% by fiscal year 2028, according to the bank’s estimates. For Alphabet, depreciation could rise from 4% of revenue in calendar year 2024 to 11% over the same period.x

Rising Leverage and Market Implications

The off-balance-sheet dimension adds further risk. Morgan Stanley noted that the prevalent use of leases is inflating true capital intensity beyond what traditional capex figures capture. Combined procurement commitments across major hyperscalers are approaching $1 trillion, while the group is expected to account for about 40% of all Russell 1000 cash capex over 2026-28, totaling more than $2 trillion.yahoo

Reuters reported Tuesday that global AI-related debt issuance is expected to reach nearly $570 billion in 2026, more than double the prior year, as tech firms tap credit markets to fund their infrastructure ambitions. The question confronting investors is whether AI revenues will scale fast enough to justify spending that now dwarfs the capital cycle that preceded the 2000 crash.reuters