Morgan Stanley Research Report Analysis: AI Infrastructure ROIC Can Reach Up to 40%, Companies Building Their Own Compute Are the Biggest Winners

By: Rita

Hyperscalers are investing hundreds of billions of dollars annually to build data centers. The market treats this money as cost. Morgan Stanley calculated a different account in its research report on July 27.

Building proprietary GPU clusters for compute leasing can yield a Return on Invested Capital (ROIC) of up to 40%. Building proprietary compute for API calls can also achieve 40%. But if compute is rented, ROIC drops to just 25%.

The gap stems from pricing power during compute shortages. Whoever holds proprietary compute controls the upstream of profit distribution. Morgan Stanley maintains an Overweight rating on Microsoft, Amazon, Meta, and Google, with target prices of $600, $330, $775, and $400 respectively.

The market sees these companies spending money. Morgan Stanley sees the money spent by these companies turning into profits. If the 40% ROIC holds, the AI businesses of these four will continue to exceed expectations, something the market has not yet priced in.

Different Compute Models, Significant Differences in ROIC Returns

Morgan Stanley breaks down the commercialization path of generative AI into three models, with ROIC ranging from 25% to 40%. The gap between models is essentially a profit distribution pattern determined by the method of compute acquisition.

·The first category is hyperscalers building proprietary GPU clusters for compute leasing. Based on Nvidia GB300, Morgan Stanley assumes 410,000 GPUs, 3.6 billion total GPU hours, 75% utilization, and $8.50 per hour rent. Incremental Earnings Before Interest and Taxes (EBIT) margin is 60% to 70%, with ROIC between 25% and 40%.

In terms of cost structure, Morgan Stanley breaks costs down into depreciation of IT equipment (servers + network) and non-IT facilities (power, cooling, racks), plus energy costs and O&M expenses. In an ecosystem where compute remains scarce, leasing prices are determined by supply and demand, not constrained by cost-plus. This is the fundamental reason why profit margins can reach 60% to 70%.

·The second category is model vendors building proprietary compute for API calls. Morgan Stanley assumes 65% of compute is used for inference, each GPU processes 2,750 tokens per second, and the charge is $1.75 per million tokens. Incremental profit margin is 70%, with ROIC around 40%.

Morgan Stanley believes this pricing level sends a positive signal to model vendors such as Google Gemini, Meta API platform, and xAI Grok. However, model vendors face a persistent constraint: compute needs to be allocated between "revenue-generating inference" and "training to maintain technological leadership." Compute invested in training generates no current revenue but bears all depreciation and operating costs. This trade-off will directly affect the sustainability of ROIC.

·The third category is renting third-party compute for APIs. Model vendors do not hold hardware and rent compute from hyperscalers by the hour. After adding a layer of "intermediary profit," incremental profit margin drops to 30%, with ROIC around 25%. In this model, the rental income of hyperscalers constitutes a cost item for model vendors, and compute pricing power lies entirely upstream.

Putting the three models together, the ROIC of proprietary compute is nearly double that of the rental model. The core of the gap lies in the ownership of pricing power in a compute-scarce ecosystem. Whoever owns proprietary compute takes the initiative in profit distribution. Morgan Stanley's ROIC calculation has deducted training costs, using a more conservative caliber than the market. Token throughput efficiency at the chip and software levels continues to improve, so the ceiling for ROIC may be higher than current assumptions.

Four Major Cloud Vendors: Breakdown of Earnings and Valuation Scenarios

Morgan Stanley maintains an Overweight rating on all four vendors, each with clear driving logic and risk-return structure.

·Microsoft Target price $600, corresponding to 25 times P/E ratio based on expected EPS of $23.86 for fiscal year 2028. The current stock price corresponds to less than 16 times GAAP EPS for fiscal year 2028; Morgan Stanley believes the valuation is low.

In terms of driving logic, Morgan Stanley values the superimposed effect of the Azure growth inflection point and Copilot monetization capability. Adoption of Azure AI services is accelerating, penetration of M365 Copilot among business customers continues to rise, and upgrades to higher-priced SKUs are lifting the average revenue per user.

Bull case target price $795, corresponding to approximately 29 times P/E ratio based on EPS of $27.39. No target price given for bear case, corresponding to approximately 12 times P/E ratio based on EPS of $21.64. In the bear case scenario, Azure growth continues to slow due to base effects, Copilot adoption is limited, and macro weakness suppresses enterprise IT spending.

·Amazon Target price $330, corresponding to 25 times P/E ratio based on the average expected EPS of approximately $13 from 2027 to 2028. Morgan Stanley believes Amazon's profit improvement comes from the combined drive of three engines: AWS cloud business growth is accelerating, advertising business continues to contribute high-margin revenue, and retail business fulfillment efficiency continues to improve.

Recurring revenue from Prime members and positive business structure switching are the core logic supporting valuation premium. Bull case $400, bear case $210.

·Meta Target price $775, based on Discounted Cash Flow model, implying approximately 23 times expected EPS for 2027. Morgan Stanley expects Meta's advertising revenue to grow by approximately 27% in 2026, with AI investment being the main driving factor.

AI is improving Reels engagement and monetization efficiency, ad measurement and attribution capabilities continue to recover after new privacy policies, and new ad products like click-to-message are opening incremental space. Bull case $1000, no specific number given for bear case.

·Google Target price $400, corresponding to approximately 24 times P/E ratio based on the average expected EPS ($15 to $18) from 2027 to 2028, equivalent to 1.6 times PEG ratio, representing a premium of approximately 35% compared to the industry median. Morgan Stanley believes AI-driven search, YouTube, and cloud platform innovations are improving the predictability of long-term growth, and new products like AI Overviews maintain ad monetization efficiency while improving user experience. Bull case $450, bear case $225.

Valuation Logic May Face Reassessment

If the 40% ROIC is real, then these hundreds of billions of dollars in capital expenditure are creating value. The market's perception of this money will change, and once it does, the valuation frameworks of the four companies must change accordingly.

Disclaimer
This article is a compilation and interpretation by TechFlow Research of a third-party brokerage research report (Morgan Stanley, July 27, 2026), combined with organized public market information. The ratings, target prices, earnings forecasts, and related judgments cited in the text are the views of the brokerage's analysts, represent only the position of their affiliated institution, do not represent the views of TechFlow Research, and do not constitute any investment advice.
The market has risks, decisions must be independent. This article should not be used as a basis for buying or selling any securities.