Morgan Stanley Provides Reassurance: "Aggressive" AI Capital Expenditure by Amazon (AMZN.US), Google (GOOGL.US), and Microsoft (MSFT.US) Will Eventually Yield Substantial Returns

According to Zhitong Finance APP, despite investor concerns about the seemingly "extreme" scale of capital expenditures by hyperscale cloud service providers—Amazon (AMZN.US), Google (GOOGL.US), and Microsoft (MSFT.US)—Morgan Stanley believes that these enormous investments will eventually yield considerable returns.

A team of analysts led by Brian Nowak pointed out that although the three major cloud giants have a combined capital expenditure of as much as $1.4 trillion, their calculations suggest that these investments could generate a return on invested capital (ROIC) of 25% to 50%.

In a detailed investor report, Nowak stated, "We are optimistic about the long-term ROIC of these investments, and for the first time, we have introduced three generative AI (GenAI) ROIC analytic frameworks and models, providing a detailed explanation of the attractive incremental ROIC (about 25% or higher) and unit economics we believe are achievable. The three business scenarios are: 1) hyperscale GPU leasing business; 2) model-enabled API business; 3) model-enabled API business running on third-party infrastructure."

Nowak pointed out: "Our underlying revenue calculations for hyperscalers involve elasticity analysis of Nvidia GB300 GPU hourly pricing and utilization rates. Major costs include depreciation of IT equipment (servers/networks) and non-IT assets (power shells), energy costs, and other operating expenses (personnel, maintenance, etc.). While GPU pricing is a key variable to be closely monitored (with rates varying across hyperscalers, chip types, reserved vs. spot/on-demand instances), overall, we estimate this business could generate an incremental EBIT profit margin of about 60%–70%, with an ROIC range of 25%–40%."

The model-enabled API businesses targeting developers, enterprises, and small and medium-sized enterprises could have ROIC potential of 40% or more.

Nowak wrote: "Here again, despite heavy upfront investments, the incremental unit economics are quite attractive. In this scenario, we adjusted the computational capacity ratio used for inference (vs. training), tokens processed per GPU per second, and token pricing (these are key levers determining profitability and ROIC)... But the importance of product innovation (driving token pricing) and chip/software innovation (driving token throughput) precisely underscores the need for large-scale investment to win the market."

Finally, the model-enabled API business operating on third-party infrastructure is expected to generate about 25% ROIC.

Nowak added: "Similar to model providers with their own capacity, token pricing and efficiency (throughput) driven by product innovation are crucial for the scalable profitability of this business. However, in this scenario, computing costs (i.e., revenue sharing with hyperscalers) also play a pivotal role in the unit economics."