The surge in AI infrastructure investment is propelling technology giants into an unprecedented cycle of heavy capital expenditure. According to a recent Morgan Stanley study, major cloud service providers, including Amazon, Alphabet, Meta Platforms, Inc., Microsoft, and Oracle, are projected to exceed the peak capital expenditure intensity seen during the dot-com bubble, signaling a structural shift in the tech industry's business model.
A report from Morgan Stanley dated February 26, 2026, forecasts that the capital expenditure-to-sales ratio for these five hyperscale cloud providers will reach 34%, 39%, and 37% in 2026, 2027, and 2028, respectively, surpassing the dot-com era peak of approximately 32%. When including financing leases, these ratios rise further to 38%, 44%, and 45%. Over the next three years, the combined capital expenditure of these companies is expected to exceed $2 trillion, accounting for about 40% of the total capital spending among Russell 1000 index constituents.
However, this explosive expansion in capital spending has not been matched by equivalent upward revisions in revenue projections. Over the past six months, consensus estimates for 2026–2027 capital expenditure have been raised by more than $630 billion, while revenue expectations have seen only limited adjustments. As a result, free cash flow projections for hyperscale cloud providers continue to decline. In contrast, semiconductor AI-enabling companies have seen their 2026 revenue consensus estimates rise by approximately 60% over the past two years, far outpacing the roughly 8% increase for hyperscalers, positioning them as the most direct financial beneficiaries of the current AI investment cycle.
Capital intensity has now exceeded the historical high of the dot-com bubble. Six months ago, Morgan Stanley characterized the AI infrastructure build-out as approaching, but not yet surpassing, the capital intensity of the dot-com era's fiber-optic expansion peak. Current projections, however, indicate that capital intensity will "far exceed" the dot-com peak of around 32%.
The report also emphasizes that measuring only traditional capital expenditure underestimates the scale of this investment cycle. Financing leases, which effectively represent debt-financed asset acquisition, should be included in total investment assessments. While financing leases were scarcely used during the dot-com era, hyperscale cloud providers are now committing to data center leases totaling hundreds of billions of dollars. Morgan Stanley's software analysts estimate that financing lease-related capital expenditure from Microsoft and Oracle alone will push the overall hyperscaler capex-to-sales ratio to 38%, 44%, and 45% for 2026–2028.
In terms of impact on the Russell 1000 index, hyperscale cloud providers accounted for more than 150% of the index's capital expenditure growth in 2025—meaning capital spending among the remaining constituents actually contracted. While hyperscalers saw a year-over-year capital expenditure increase of about 70%, the rest of the index declined by 6%. Morgan Stanley expects hyperscalers to represent approximately 40% of the Russell 1000's total capital expenditure by 2026, doubling from 2024 levels, and potentially rising to 49% by 2028.
Capital expenditure revisions have reached record levels, while revenue forecasts lag significantly. Since September 2025, consensus estimates for hyperscaler capital expenditure in 2026 and 2027 have been revised upward by about 1.5 times, with Morgan Stanley's own analyst projections raised by nearly 1.8 times. On a company-specific basis, Alphabet's 2026 capital expenditure consensus has been raised by 117% compared to a year ago, Meta Platforms, Inc. by 96%, Amazon by 75%, and Oracle by 264%. These adjustments have been "step-like" rather than gradual, reflecting the difficulty in predicting this investment cycle as companies continuously update data center expansion plans and compete to secure key supply chain components.
In stark contrast to the rapid upward revisions in capital expenditure, revenue forecasts have remained largely stagnant, leading to downward adjustments in free cash flow expectations. Over the past year, combined capital expenditure projections for the five companies for 2026 have been raised by more than $310 billion, while revenue revisions totaled only about $130 billion. As fixed-cost bases expand, these companies' operating leverage will increase, making future earnings and free cash flow more sensitive to changes in revenue expectations.
The widespread use of financing leases has significantly amplified actual investment scale. As of the latest earnings reports, the five companies have committed to future lease obligations exceeding $660 billion, with Oracle at approximately $248 billion, Microsoft at $155 billion, Meta Platforms, Inc. at $104 billion, Amazon at $96 billion, and Alphabet at $59 billion. Notably, Alphabet's lease commitments have increased roughly seven-fold since 2024, while Meta Platforms, Inc.'s have risen by over 200%.
Financing leases have a pronounced impact on individual companies' capital intensity. For Microsoft, traditional capital expenditure alone would result in a capex-to-sales ratio of about 29% in fiscal 2026 and 2027, but including financing leases raises this to approximately 43% and 42%, respectively. Oracle's situation is more extreme—the company is acquiring all data center shells via leases. Under traditional measures, Oracle's capex-to-sales is projected at 75% and 119% for fiscal 2026 and 2027, but including financing leases pushes these ratios to 107% and 201%, meaning total reinvestment would exceed total revenue in both fiscal years.
While capital expenditure is highly concentrated among hyperscale cloud providers, semiconductor AI-enabling companies are the clearest near-term financial beneficiaries. This divergence stems from differences in revenue certainty: hyperscalers' large-scale advance purchases of GPUs and other chip components provide semiconductor suppliers with visible near-term revenue, while the cloud providers themselves must monetize these computing assets over several years through large language model commercialization, sustained computing demand, and product differentiation—a process with higher uncertainty.
Market performance reflects this logic. Since December 2023, North American semiconductor AI-enabling stocks have outperformed hyperscale cloud providers and the broader AI-enabled sector by 272% and 224%, respectively. The market is currently valuing semiconductor companies based on their near-term earnings visibility, while adopting a wait-and-see approach toward revenue realization paths for hyperscalers and other AI-enabled segments.
Morgan Stanley analyst Brian Nowak suggests that Meta Platforms, Inc., Alphabet, and Amazon are leveraging AI investments, data accumulation, and scale to accelerate user engagement and monetization. Keith Weiss views Oracle's data center expansion as a potential revenue opportunity but notes it requires substantial funding. The ongoing trend of capital expenditure upward revisions will also lead to rising depreciation expenses, which, without corresponding revenue increases, will exert significant pressure on profit margins.
Comments