Morgan Stanley's latest analysis offers a measured perspective on whether the massive AI investments by tech giants are yielding adequate returns. The investment bank's calculations show that model companies offering API services through proprietary computing infrastructure can achieve a return on invested capital (ROIC) of up to 46%. Meanwhile, hyperscale cloud providers renting out GPUs see an ROIC of approximately 31%, and even model companies relying on third-party infrastructure can attain an ROIC of roughly 25%. These figures suggest the trillion-dollar AI arms race represents more than just unchecked capital expenditure.

What is particularly noteworthy is that AI investments are increasingly demonstrating self-sustaining financial capability. Morgan Stanley projects that the combined operating cash flow of the four major tech behemoths—Amazon, Alphabet, Microsoft, and Meta Platforms, Inc.—will climb from $739 billion in 2026 to $1.23 trillion by 2028. Over the same period, the need for new debt financing is expected to decline from $238 billion to $90 billion. By 2028, these companies' incremental debt requirements would represent only about 7% of their operating cash flow, suggesting that financial strain is not worsening in tandem with swelling capital expenditures.

However, the breakneck pace of capital spending may be approaching a cyclical peak. Capital expenditures on data centers by the four hyperscale cloud providers are projected to rise from $917 billion in 2026 to $1.47 trillion in 2027—a roughly 60% year-over-year increase—before growth decelerates sharply to just 12% in 2028. This signals a shift in market focus from building ever-larger computing capacity to converting existing infrastructure into tangible revenue and profits.

This is precisely where Morgan Stanley sees opportunity. As capital expenditure growth slows and AI application adoption accelerates, investment dollars are likely to rotate from hardware, semiconductors, and memory toward models, cloud platforms, and software applications. The "ROI examination" for AI spending is transitioning from the construction phase into the commercialization and monetization stage.

Capital Expenditure: Peaking in 2027, Decelerating in 2028

According to Morgan Stanley's projections, data center capital expenditures by hyperscale cloud providers will escalate from roughly $466 billion in 2025 to $917 billion in 2026, reaching approximately $1.47 trillion in 2027 and $1.64 trillion in 2028. However, the growth rate is expected to plummet from about 60% in 2027 to merely 12% in 2028.

Among the major players, Alphabet is pursuing the most aggressive expansion strategy, with data center capital expenditures forecast to grow 83% year-over-year in 2027. Amazon, Meta Platforms, Inc., and Microsoft are projected to see growth rates of approximately 50%, 55%, and 43%, respectively, though all are expected to decelerate notably by 2028.

Morgan Stanley points to real-world constraints—including chip availability, rack infrastructure, land, power supply, and labor—that are limiting further expansion. Additionally, the giants have already "pre-built" substantial data center capacity to accommodate demand through 2027 to 2029, leaving limited room for further front-loading of capital expenditure. The pivotal shift in 2028 is not about AI demand reaching a ceiling but rather about infrastructure investment transitioning from rapid expansion into a digestion phase.

Compute Capacity: Nearly Quadrupling in Three Years

Even as capital expenditure growth moderates, total compute capacity will continue to expand substantially. Morgan Stanley estimates that the aggregate compute capacity of the four hyperscale cloud providers will surge from approximately 36 gigawatts in 2025 to around 144 gigawatts by 2028—nearly a fourfold increase.

Alphabet is expected to be among the largest contributors to new capacity, adding roughly 9 GW and 11 GW in 2027 and 2028, respectively. This expansion is primarily designated for training Gemini, driving Google Cloud Platform growth, and supporting generative AI features across Search and YouTube.

The composition of compute capacity is also evolving. Morgan Stanley projects that custom ASICs will grow from 34% of new compute capacity in 2025 to 66% by 2028, with Alphabet's TPUs and Amazon's Trainium chips serving as primary growth drivers. While demand for AI compute continues to rise, the industry is pivoting from simply acquiring GPUs to enhancing unit-level compute efficiency. NVIDIA retains its dominant position, but the strategic importance of cloud providers' in-house silicon is steadily increasing.

AI Returns: Proprietary Infrastructure with Model Layer Yields Highest Returns

Morgan Stanley has calculated the return on invested capital for three distinct generative AI business models. Under the IaaS model where hyperscale cloud providers rent out GPUs, using GB300 as the benchmark, each gigawatt corresponds to approximately $23 billion in revenue, generating an ROIC of around 31%.

The most lucrative model involves model companies using proprietary infrastructure to offer API services, generating roughly $30.4 billion in revenue per GW with an after-tax operating profit of about $17.9 billion, translating to an ROIC of approximately 46%. In contrast, model companies reliant on third-party infrastructure for API delivery can generate higher top-line revenue of about $40.5 billion per GW, but after accounting for compute leasing costs, the ROIC drops to only around 25%.

This analysis suggests that the most attractive segments of the AI value chain are not necessarily those simply renting out compute power, but rather companies that control model capabilities, infrastructure, and commercialization expertise. Morgan Stanley identifies Meta Platforms, Inc. and Alphabet as prime examples of this winning formula.

Market Potential: AI Commercialization Is the Next Critical Phase

Morgan Stanley estimates the global generative AI total addressable market at $50 trillion to $60 trillion, encompassing a knowledge work segment of approximately $20 trillion to $30 trillion and a consumer segment of around $30 trillion.

Drawing parallels to the public cloud adoption curve, enterprise AI spending is projected to reach approximately $812 billion by 2027, representing penetration of only about 4% of the knowledge work TAM. Morgan Stanley believes AI diffusion may outpace public cloud adoption, primarily because enterprises can avoid large-scale infrastructure migration and AI delivers quantifiable productivity gains more rapidly.

Significantly, concrete signs of commercialization have begun to materialize. As of the second quarter of 2026, roughly 25% of S&P 500 constituent companies were able to quantify benefits derived from generative AI—a notable increase from the 14% figure recorded just one year earlier.

This indicates that AI investment is transitioning from the "build first, await demand" approach into a self-reinforcing cycle of compute investment, application growth, and revenue realization. Whether the application layer can consistently contribute revenue and profits will become the central driver of valuation in the next phase.