AI Capex Intensity to Surpass Internet Bubble Peak
AI summary card
AI Capex Intensity to Surpass Internet Bubble Peak
Morgan Stanley predicts that hyperscalers' capex-to-sales ratio will reach 36%-44% in 2026-28, exceeding the peak during the internet bubble; massive lease and purchase commitments support construction, but significant depreciation pressure is expected over the next three years.
- Capex/Sales ratio projected at 36%/44%/42% for 2026-28, far exceeding the 32% peak of the internet bubble era
- AI-related spending is expected to account for over 50% of total Russell 1000 Index capex in 2026
- RPO balances of major AI compute players have nearly tripled in the past year, surpassing $2 trillion
- Long-term purchase commitments approach $1 trillion, with lease commitments exceeding $800 billion
- Cumulative depreciation for MSFT, ORCL, META, and GOOGL is projected to exceed $520 billion over the next three years
- Finance leases elevate actual capital intensity, with some companies' lease-inclusive Capex/Sales potentially exceeding 100%
Report interpretation
Overview
This report is Morgan Stanley's latest chart update on key financial metrics of the AI ecosystem. The core conclusion is that current AI infrastructure construction is undergoing a record-breaking capital cycle. The report notes that hyperscalers' capital expenditure intensity is poised to surpass the historical peak of the 2000 internet bubble, with this round of investment heavily reliant on off-balance-sheet financing tools such as finance leases and long-term purchase commitments. While the surge in Remaining Performance Obligations (RPO) validates the authenticity of demand, the ensuing massive depreciation will be a key variable suppressing margins in the coming years.
Core views
Record-High Capital Expenditure Intensity: The report forecasts that hyperscalers' capex-to-sales ratios will reach 36%, 44%, and 42% from 2026 to 2028, significantly higher than the 32% peak during the fiber-optic build-out phase of the internet bubble. Including finance leases, actual capital intensity is even higher; for instance, Oracle's lease-inclusive Capex/Sales ratio could soar to 189% in FY27, while Microsoft's could reach 64%. AI has become the absolute investment theme, with AI-related spending expected to account for over 50% of the Russell 1000 Index's total capital expenditure in 2026, with hyperscalers contributing approximately 90%. Off-Balance-Sheet Leverage and Financing Innovation Support Construction: To meet record investment demands, companies are extensively using off-balance-sheet instruments. As of Q1 2026, total long-term purchase commitments by major vendors approached $1 trillion, with uncommenced lease commitments exceeding $800 billion. These commitments do not appear on the balance sheet before delivery but constitute substantial operating leverage. Meanwhile, unpaid capex within accounts payable has reached approximately $110 billion, leading to a significant extension in Days Payable Outstanding (DPO) compared to historical averages, reflecting a widening gap between AI monetization and payments to suppliers. Depreciation Becomes the Next Margin Focus: As capital investments convert into fixed assets, depreciation expenses will rise sharply. The report estimates that cumulative depreciation for Microsoft, Oracle, Meta, and Alphabet over the next three years will exceed $520 billion. For example, Oracle's depreciation as a percentage of revenue is projected to rise from 7% in FY26 to 28% in FY28; Meta's is expected to rise from 9% to 19% over the same period. Unless non-depreciation costs decline or revenue grows beyond expectations, margins will face severe tests. However, the rapid growth in Construction in Progress (CIP) balances (Oracle's grew by approximately 200% over the past year) indicates a lag in the conversion of capital expenditures into income statement expenses, temporarily mitigating the direct impact on earnings. Demand Validation and Extended Contract Terms: The explosive growth in Remaining Performance Obligations (RPO) provides strong evidence of AI demand. Total RPO for major AI compute players has nearly tripled in the past year, breaking through $2 trillion. Oracle's RPO grew by over 300% year-over-year, while Microsoft and Alphabet grew by 97% and 406%, respectively. Notably, the weighted average term of RPOs is lengthening, indicating that customers are locking in compute resources through longer-term contracts, thereby supporting future revenue visibility.
Analysis framework
The report adopts an 'Accounting and Valuation' perspective, constructing a comprehensive view of cash flows in the AI ecosystem by analyzing hard data from SEC filings, including capital expenditures, lease commitments, purchase obligations, construction in progress, and remaining performance obligations. The analytical framework follows the logical chain of 'Capital Investment → Financing Structure → Asset Conversion → Profit Impact': First, it benchmarks investment heat against historical cycles using the Capex/Sales ratio; second, it deconstructs off-balance-sheet leases and purchase commitments to reveal true leverage levels; third, it predicts future margin pressure through CIP and depreciation models; finally, it validates the sustainability and quality of demand using RPO and its term structure. This approach moves beyond pure technical narratives, quantifying the cost and sustainability of the AI boom from the perspective of financial articulation.
Methodology notes
Off-Balance-Sheet Commitment Restoration Analysis
The report looks beyond debt on the balance sheet, focusing on tracking off-balance-sheet items such as 'uncommenced leases' and 'purchase commitments.' These items do not appear as liabilities before delivery but represent future rigid cash outflows. By comparing off-balance-sheet commitments with operating cash flow, one can more accurately assess the actual financial leverage and risk exposure behind AI construction.
Cross-Cycle Capital Intensity Benchmarking
It horizontally compares the current AI cycle's Capex/Sales ratio with the peak of the 2000 internet bubble to determine the historical position of this equipment investment cycle. This cross-cycle benchmarking helps identify whether the current phase represents irrational exuberance or structurally driven expansion fueled by genuine demand.
Construction in Progress (CIP) as a Depreciation Buffer
The report monitors changes in CIP balances, viewing them as a 'time valve' for the conversion of capital expenditures into depreciation expenses. A significant increase in CIP implies that substantial funds have been invested but not yet capitalized, explaining why current massive capex has not immediately impacted the income statement, while also signaling the timing of future concentrated depreciation expense releases.
RPO Term Structure Analysis
It examines not only the total growth of Remaining Performance Obligations (RPO) but also decomposes their term composition (e.g., within 12 months vs. beyond 12 months). Lengthening RPO terms typically indicate that customers are willing to sign longer contracts to lock in scarce compute capacity, serving as a leading indicator of demand rigidity and industry supply tightness, reflecting supply-demand dynamics more effectively than mere order values.
Key data
- 2026-28E Capex/Sales Ratio36% / 44% / 42%Far exceeds the 32% peak of the internet bubble era; figures are higher when including finance leases
- AI-Related Capex Share (2026E)>50%Share of total Russell 1000 Index capex, with hyperscalers contributing approximately 90%
- Total RPO of Major AI Players>$2tnNearly tripled within the past year, reflecting a surge in long-term contract signings
- Total Long-Term Purchase Commitments~$1tnOff-balance-sheet obligations prior to delivery, supporting supply chain capacity expansion
- Total Uncommenced Lease Commitments>$800bnFinance leases elevate actual capital intensity, increasing off-balance-sheet operating leverage
- Projected Cumulative Depreciation for Big Four Over Three Years>$520bnCovers MSFT, ORCL, META, GOOGL; will be a core source of margin pressure
- Oracle RPO YoY Growth Rate>300%Highest growth rate; Microsoft and Alphabet grew by 97% and 406%, respectively
Impact & implications
For the AI ecosystem, record capital expenditures and off-balance-sheet commitments imply that infrastructure construction is still accelerating, providing upstream equipment manufacturers and compute providers with strong order security in the medium term. However, as construction in progress gradually converts to fixed assets, massive depreciation will substantially squeeze hyperscalers' margins. Future ability to maintain profit expectations will heavily depend on the speed of AI application monetization and revenue growth elasticity. Furthermore, the widespread use of finance leases and purchase commitments increases implicit corporate leverage; if AI commercialization returns fall short of expectations, these off-balance-sheet obligations may transform into financial risks.
Risks
- AI commercialization monetization speed falls short of expectations, causing massive depreciation and off-balance-sheet commitments to become financial burdens
- Off-balance-sheet leverage constituted by finance leases and purchase commitments may trigger liquidity pressures during demand downturns
- Continuous upward revisions in capital expenditures outpacing revenue upgrades may lead to long-term pressure on free cash flow
What to watch
- Quarterly depreciation expense changes and margin guidance for hyperscalers
- Pace of Construction in Progress (CIP) capitalization relative to new capex additions
- Changes in RPO term structure and trends in long-term contract signings
- Scale variations in off-balance-sheet leases and purchase commitments, along with adjustments to financing terms