Morgan Stanley: AI Infrastructure ROIC Reaches Up to 46%; Cloud Vendors Maintain Overweight Rating

According to Chaoxiang Research, Morgan Stanley released a report on July 27, quantifying the incremental return on invested capital (ROIC) for generative AI investments for the first time. The report constructed three valuation frameworks, estimating ROIC at approximately 31% for hyperscale cloud providers’ GPU leasing services, 46% for proprietary infrastructure-based API services, and 25% for third-party compute API services. Under baseline assumptions, a 1-gigawatt (GW) data center would be configured with approximately 410,000 NVIDIA GB300 GPUs, operating at 75% utilization, with an hourly rental price of $8.50. Combined capital expenditures by the three major cloud giants are projected to exceed $1.4 trillion. Morgan Stanley maintains overweight ratings on Microsoft, Amazon, Meta, and Google. The report notes that as AI transitions from training to inference, demand for GPU compute power will continue to rise. Companies building their own infrastructure will benefit from pricing power in an environment of compute scarcity, generating substantial profits. If Morgan Stanley’s estimates hold true, these hundreds of billions of dollars in AI capital spending will shift from being perceived as “costs” to being recognized as “growth assets.”