Morgan Stanley: AI infrastructure ROIC can reach up to 46%, and mega-cap vendors maintain an Overweight rating

Deep Tide TechFlow message, according to trend-and-direction research, Morgan Stanley published a research note on July 27, for the first time quantifying the incremental capital return on investment (ROIC) for generative AI investment. The report builds three sets of calculation frameworks: ROIC for the GPU leasing business of hyperscale cloud service providers is about 31%, ROIC for the owned infrastructure model API business is about 46%, and ROIC for the third-party compute API business is about 25%. Under baseline assumptions, a single 1 gigawatt (GW) data center would be configured with about 410k NVIDIA GB300 GPUs, utilization at 75%, and an hourly rental price of $8.5. Total combined capital expenditures across the three major cloud giants are expected to exceed $1.4 trillion.

Morgan Stanley has kept over-weight ratings on Microsoft, Amazon, Meta, and Google. The report points out that as AI moves from the training stage into the inference stage, demand for GPU compute power will continue to grow. With pricing power in an ecosystem where compute supply is constrained, in-house compute providers are set to capture significant profits. If Morgan Stanley’s calculations hold, these several hundred billion dollars in AI capital expenditures will shift from a “cost” mindset to a “growth asset” mindset.