AI Capex Has Officially Surpassed the Dot-Com Era's Peak. What Comes Next?

The AI infrastructure buildout has entered unprecedented territory. Morgan Stanley’s latest research reveals that the capital expenditure intensity of the five largest hyperscalers—Amazon, Alphabet, Meta, Microsoft, and Oracle—has not just approached but decisively surpassed the peak levels seen during the dot-com bubble. The numbers are staggering, the revisions relentless, and the financial implications extend far beyond the technology sector itself.

Capex-to-Sales Ratios: How AI Spending Compares to the Dot-Com Era

Morgan Stanley’s Global Valuation, Accounting & Tax strategists now forecast that the cash capex-to-sales ratio for the five hyperscalers will reach 36% in 2026, 44% in 2027, and 42% in 2028 . The 44% figure for 2027 is well above the approximately 32% peak recorded during the dot-com era . When finance leases are factored in, these ratios are pushed even higher .

This is a dramatic escalation from earlier estimates. In September 2025, Morgan Stanley had projected a 26% capex-to-revenue ratio by 2027—a level that merely approached the dot-com peak . The upward revision underscores how rapidly the AI spending race has intensified.

For context, the current intensity also surpasses the 20% ratio seen during the shale oil boom . Morgan Stanley’s analysis notes that just a handful of U.S. tech companies are now set to spend almost as much on capital equipment as the entire U.S. economy .

Combined Capex Forecasts: From $450 Billion to $1.16 Trillion

The trajectory of Morgan Stanley’s capex forecasts tells its own story. A year ago, the bank projected combined hyperscaler capex at roughly $450 billion for both 2026 and 2027 . Those estimates have now more than doubled:

• 2026: Approximately $800 billion (with some estimates at $805 billion)
• 2027: Approximately $1.16 trillion

The 2027 figure is more than four times what the group spent in 2024 . After rising 70% year-over-year in 2025, hyperscaler capex is poised to continue accelerating, with an incremental $2 trillion expected through 2028 . Morgan Stanley’s equity analysts led by Brian Nowak now model these dramatically higher figures following first-quarter earnings reports .

The Depreciation Wave: $680 Billion in Charges Ahead

The enormous capital outlays are creating a looming depreciation problem. Morgan Stanley provides two overlapping estimates:

• Across all five hyperscalers, cumulative depreciation over the next three years (2026–2028) is expected to exceed $520 billion .
• For Microsoft, Oracle, Meta, and Alphabet specifically, total depreciation charges over the next four years (2026–2029) could top $680 billion .

The company-level projections are particularly striking. Alphabet’s depreciation expenses could increase fourfold by the end of 2028 . Oracle’s depreciation charge of roughly $4 billion in 2025 might balloon to as much as $56 billion by 2029, which would equate to approximately 28% of consensus revenue estimates .

These charges matter because they flow directly through the income statement, compressing margins and reducing reported earnings. The Financial Times has characterized this as Big Tech’s “buy-now-book-later problem” .

Oracle Faces the Most Acute Pressure

Among the five hyperscalers, Oracle is most exposed. Its depreciation-to-revenue ratio is projected to rise from approximately 7% to 28% over the coming years . Because Oracle’s revenue base is smaller than its peers, the same dollar amount of infrastructure depreciation claims a much larger share of its top line. Morgan Stanley analysts have flagged Oracle as the company where the mismatch between heavy AI infrastructure spending and revenues will be most visible in financial statements .

Off-Balance-Sheet Risk: Hidden Leverage and SPVs

The reported capex figures may understate the true financial exposure. Morgan Stanley notes that finance leases are pushing effective capex-to-sales ratios even higher than cash capex figures suggest . More significantly, technology companies have moved more than $120 billion in data center spending off their balance sheets through special-purpose vehicles (SPVs) and similar structures in under two years, creating hidden leverage and litigation risk .

Moody’s Ratings has separately flagged that the top five U.S. hyperscalers have accumulated $662 billion in future data center lease commitments that have not yet commenced and therefore sit entirely off their balance sheets under current accounting rules . This off-balance-sheet financing has drawn scrutiny from U.S. senators, who have specifically called out structures where external investors fund and own data centers that are then leased back to the technology companies .

Debt Markets: Financing a $1.5 Trillion Gap

The hyperscalers cannot fund this buildout from operating cash flow alone. Morgan Stanley estimates that roughly $2.9 trillion in global data center capex will be required through 2028, with a $1.5 trillion financing gap that must be filled by external capital . The projected split:

• $800 billion in private credit
• $200 billion in corporate bonds
• $150 billion in securitized credit (ABS/CMBS)
• Remainder from other sources including private equity, venture capital, and bank lending

Debt issuance is already surging. Hyperscalers issued more than $100 billion of investment-grade debt in 2025 alone to finance data center buildouts . Morgan Stanley expects net debt supply from this cohort to increase by roughly 30% to 50% in 2026, potentially reaching $130 billion to $150 billion . Reuters reported that annual debt issuance tied to AI and data centers rose from $166 billion in 2023 to $625 billion in 2025 . J.P. Morgan has estimated total new debt from Big Tech at $455 billion across 27 issuers, including $357 billion in direct borrowing by the five hyperscalers .

Historical Parallels and Market Warnings

Morgan Stanley has drawn several sobering historical comparisons:

• AI-related hyperscaler capex is now set to exceed dot-com era telecom capex in both magnitude and length .
• The spending intensity surpasses not only the dot-com bubble peak but also the levels seen during the shale oil boom .
• The five hyperscalers are expected to account for roughly 40% of total Russell 1000 cash capex from 2026 to 2028, representing more than $2 trillion in spending .

Morgan Stanley Wealth Management’s chief investment officer Lisa Shalett has warned of “cracks” forming in the AI capex boom, calling the sustainability of GenAI investment the most important question for investors in the coming year. She noted: “When it comes to market-discounting scenarios, we believe we are closer to the seventh than the first” .

The broader research context underscores the scale: Morgan Stanley estimates that nearly $3 trillion of AI-related infrastructure investment will flow through the global economy by 2028, with more than 80% of that spending still ahead . The bank characterizes the buildout as more akin to an industrial build-out than speculative tech spending, but the financial engineering, depreciation burden, and debt dependency introduce risks that echo previous boom-and-bust cycles in uncomfortable ways.