More intense than the internet bubble! Tech giants gamble $2 trillion on AI, unprecedented capital strength

The wave of AI infrastructure investment is pushing tech giants into an unprecedented heavy-asset cycle. Morgan Stanley’s latest research shows that hyperscalers, represented by Amazon, Google, Meta, Microsoft, and Oracle, are expected to surpass the historical peak of internet bubble-era capital expenditure intensity, indicating a structural shift in the business models of the tech industry.

According to a report released by Morgan Stanley on February 26, 2026, the capital expenditure-to-sales ratio (capex-to-sales) for these five hyperscalers is projected to reach 34%, 39%, and 37% in 2026 to 2028, respectively, exceeding the peak of about 32% during the internet bubble era.

Including financing leases, this ratio could further rise to 38%, 44%, and 45%. Meanwhile, the total capital expenditure of these companies over the next three years will exceed $2 trillion, accounting for roughly 40% of the total capital expenditure of the Russell 1000 index components.

However, the explosive growth in capital expenditure has not led to a corresponding increase in revenue estimates. Morgan Stanley notes that over the past six months, market consensus expectations for capital spending in 2026-2027 have been collectively raised by over $630 billion, but revenue expectations have been revised upward by a much smaller margin, leading to continued downward revisions in free cash flow (FCF) forecasts for hyperscalers. In contrast, semiconductor AI-enabled companies have seen their 2026 revenue consensus expectations increase by about 60% over the past two years—far surpassing the roughly 8% increase for hyperscalers—making them the most directly financially benefited group in this AI investment cycle.

Capital intensity surpasses internet bubble highs

Morgan Stanley’s report states that six months ago, the AI build-out wave was characterized as “approaching but not yet surpassing” the fiber-optic construction peak of the internet bubble in terms of capital intensity. The latest forecast now indicates that capital intensity will “far exceed” the approximately 32% peak during the internet bubble, with capex-to-sales ratios projected at 34%, 39%, and 37% for 2026-2028.

The report also emphasizes that measuring solely by traditional capital expenditure underestimates the scale of this investment cycle. Financing leases are essentially debt-based asset acquisitions and should be included in total investment assessments. During the internet bubble era, leasing was rarely used, but now hyperscalers are signing data center lease agreements worth hundreds of billions of dollars. Morgan Stanley’s software industry analysts estimate that just Microsoft and Oracle’s leasing capital expenditures alone could push the overall hyperscalers’ capex-to-sales ratios to 38%, 44%, and 45% in 2026-2028.

Regarding the Russell 1000 index, hyperscalers contributed over 150% of the index’s capital expenditure increase in 2025—meaning the other components’ capital expenditures are effectively shrinking. Hyperscalers’ capital expenditure grew by about 70% year-over-year, while the rest of the index declined by 6%. Morgan Stanley expects that by 2026, hyperscalers will account for about 40% of the Russell 1000’s total capital expenditure—doubling from 2024—and may further rise to 49% by 2028.

Record-breaking capital expenditure revisions lag behind revenue forecasts

A notable feature of this investment cycle is the unprecedented speed and magnitude of upward revisions in capital expenditure forecasts. Since September 2025, market consensus expectations for hyperscalers’ 2026 and 2027 capital spending have each been raised approximately 1.5 times, with Morgan Stanley analysts’ own forecasts increasing by about 1.8 times.

Looking at individual companies, Google’s 2026 capex-to-sales consensus has been revised upward by 117% compared to a year ago, Meta by 96%, Amazon by 75%, and Oracle by as much as 264%. Morgan Stanley analyst Todd Castagno’s team notes that these revisions are “stepwise” rather than gradual, indicating that this cycle is highly unpredictable—management teams are continuously updating data center expansion plans, and companies are competing to lock in key supply chains, further complicating forecasts.

In stark contrast, revenue revisions have remained nearly flat, while FCF expectations have declined. Data shows that over the past year, the five companies’ combined capital expenditure forecasts have been raised by over $310 billion, while revenue revisions total only about $130 billion. Morgan Stanley points out that as fixed cost bases expand, operational leverage will increase, making future profits and FCF more sensitive to revenue changes.

Leasing significantly amplifies actual investment scale

Recently, hyperscalers have greatly expanded their use of financing leases, further elevating actual capital intensity levels. According to the latest financial reports, these five companies have committed over $660 billion in future lease obligations, including Oracle (~$248 billion), Microsoft (~$155 billion), Meta (~$104 billion), Amazon (~$96 billion), and Google (~$59 billion). Notably, Google’s lease commitments have increased about sevenfold since 2024, and Meta’s have grown over 200% in the same period.

Leasing has a particularly pronounced impact on individual companies’ capital intensity. For example, in Microsoft’s case, if only traditional capex is considered, its FY26 and FY27 capex-to-sales ratios are about 29%. Including leasing, these ratios jump to approximately 43% and 42%. Oracle’s situation is even more extreme—Oracle is leasing all data center shells, and under traditional accounting, its FY26 and FY27 capex-to-sales are projected at 75% and 119%, respectively. When leasing is included, these figures rise to 107% and 201%, meaning the total reinvestment in those years exceeds the company’s annual revenue.

Semiconductors are the biggest winners, but hyperscalers still need to prove returns

Despite the heavy concentration of capital expenditure among hyperscalers, the clearest recent financial beneficiaries are semiconductor AI-enabled companies.

The fundamental reason for this divergence lies in revenue certainty: hyperscalers have made large-scale early purchases of GPUs and other chips, providing them with near-term visible revenue streams for chip suppliers; meanwhile, hyperscalers themselves need to monetize large language models, sustained computing demand, and product differentiation over the coming years, with higher uncertainty.

Market performance also reflects this logical split. Since December 2023, North American semiconductor AI-enabled companies’ stock prices have increased by 272% and 224%, respectively, outpacing hyperscalers and the broader AI-enabled sector. The market is currently willing to pay a premium for the near-term profits of semiconductor companies, while remaining cautious about revenue realization for hyperscalers and the wider AI sector.

Morgan Stanley analyst Brian Nowak believes that Meta, Google, and Amazon are accelerating user engagement and commercial monetization through AI investments, data accumulation, and scale advantages; Keith Weiss characterizes Oracle’s data center expansion as a potential revenue opportunity but emphasizes that it requires substantial capital support. The upward revision trend in capex will also lead to rising depreciation expenses, which, without corresponding revenue upgrades, will exert significant pressure on profit margins.

This insightful content is brought to you by WindTrader.

For more detailed analysis, real-time insights, and frontline research, please join the 【**WindTrader Annual Membership**】.

Risk Disclaimer and Terms of Use

Market risks exist; invest cautiously. This article does not constitute personal investment advice and does not consider individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions herein are suitable for their particular circumstances. Investment is at your own risk.