U.S. technology giants are engaged in an unprecedented AI infrastructure arms race, with capital expenditure intensity approaching peak levels seen during the internet bubble.

Latest research from Bank of America and Morgan Stanley shows that markets are severely underestimating the true scale of current AI investments while being unprepared for the impact of future depreciation expenses. Supply-demand imbalances could trigger a cloud services price war as early as 2027.

Morgan Stanley's research indicates that "hyperscale" players including Amazon.com, Alphabet, Meta, Microsoft and Oracle are expected to see their capital expenditure as a percentage of sales reach 26% by 2027, approaching the 32% peak level of the internet bubble era and exceeding the 20% seen during the shale oil boom. More critically, these public capital expenditure figures don't fully reflect the complete investment picture, as off-balance sheet tools like financing leases are increasingly being used to accelerate data center expansion, leading to underestimation of current real investment scale.

Bank of America's analysis focuses on the long-term impact of these investments. The research shows that markets are generally underestimating future depreciation costs. By 2027, market predictions for depreciation expenses for just Alphabet, Amazon.com and Meta could be nearly $16.4 billion lower than actual levels. Bank of America also states that if supply growth continues to exceed demand, the industry could see more aggressive pricing strategies emerge as early as 2027.

Capital Expenditure Race: An "Arms Race" with Underestimated Scale

Morgan Stanley's report compares the current AI investment wave to two historical capital frenzies: telecommunications fiber optic construction during the internet bubble and energy sector drilling during the shale revolution. The report notes that current capital intensity is approaching the peak of the former. Unlike previous periods, tech giants are using increasingly complex financial instruments to accelerate expansion, making traditional capital expenditure data unable to fully capture the complete investment picture.

Morgan Stanley emphasizes two factors leading to underestimation of actual investment scale:

First is the rise of financing leases. Companies like Microsoft and Oracle are increasingly using financing leases to build data centers. While economically similar to debt-financed asset purchases, initial investments typically aren't included in traditional capital expenditures, thus bypassing cash flow statements. The report finds that Microsoft and Oracle's capital intensity jumps significantly when financing leases are included. For example, according to Morgan Stanley's estimates, Microsoft's capital expenditure-to-sales ratio for fiscal 2026 will jump from 28% to 38%, while Oracle's will surge from 41% to 58%. Additionally, these giants have signed lease commitments exceeding $335 billion that haven't yet begun, indicating this trend will continue.

Second is the delayed effect of "construction in progress": Massive investments are accumulating on balance sheets as "Construction in Progress (CIP)." These assets don't incur depreciation until formally put into use, so their costs haven't yet impacted profit statements. Morgan Stanley's data shows that construction in progress balances for Alphabet, Amazon.com, Meta and Oracle all showed sharp increases over the past year, with Amazon.com growing approximately 60% ($17 billion) and Alphabet growing about 40% ($15 billion). This means substantial capital has already been spent, but its impact on profitability is just beginning.

Financial Statement "Time Bomb": Wall Street Underestimates Future Depreciation Costs

If Morgan Stanley revealed the "below the iceberg" investment scale, then Bank of America clarified how these investments will translate into real cost pressures. Its core point is that Wall Street is "slow to react" to future depreciation expense growth rates.

Bank of America analyst Justin Post noted in the report that as Alphabet, Meta and Amazon.com's combined capital expenditures increase 56% and 63% respectively in 2024 and 2025, their depreciation and amortization (D&A) expenses will inevitably accelerate growth in 2026 and beyond. Data shows significant gaps between Bank of America's 2027 depreciation expense predictions for the three giants versus market consensus:

- Alphabet: Gap of approximately $7 billion - Amazon.com: Gap of approximately $5.9 billion - Meta: Gap of approximately $3.5 billion

A total "expectation gap" of nearly $16.4 billion means these companies' future actual profitability could be far below current market consensus.

The report also points to another factor exacerbating depreciation risk: the "short-life" problem of AI assets.

Unlike traditional servers, hardware used for AI computing like GPUs faces faster technological iteration and higher workloads, with effective useful lives potentially only three to five years.

Bank of America notes that Amazon.com already shortened the estimated useful life of some servers and network equipment from six years to five years in Q1 2025, citing accelerated technological development in AI and machine learning. This reverses the trend of recent years where tech giants generally extended equipment lifespans to smooth expenses. If this trend reverses, it will lead to accelerated depreciation recognition, impacting short-term profitability.

Risk and Return: Price War Could Emerge as Early as 2027

Bank of America warns that the AI infrastructure market could repeat historical patterns where aggressive investment led to overcapacity and pricing pressure. As major tech companies continue accelerating AI infrastructure investment, there's risk of overbuilding, where computing capacity supply exceeds demand for high-value AI services.

Additionally, increasingly consistent large language model performance could weaken product differentiation, leading to infrastructure service commoditization. Meta is building multiple gigawatt-scale data centers expected to come online 2026-2029; Oracle and OpenAI's proposed $500 billion Stargate project is expected to bring substantial AI capacity in 2028-2029. If demand can't keep up with supply deployment scale, hyperscale vendors may resort to aggressive pricing strategies to maintain utilization rates, thereby compressing profit margins.

Bank of America believes that if supply exceeds consumption (which in their view could happen as early as 2027), hyperscale vendors may adopt more aggressive pricing strategies to maintain utilization, thus eroding profitability.