Morgan Stanley estimates incremental ROIC for three GPU monetization models at
roughly 31% for GPU leasing, 46% for running model APIs on owned compute, and
25% for renting third‑party compute. These figures refer to incremental returns
on a 1 GW GB300 capacity once in stable operation, not company-level returns.
Returns depend on GPU utilization, pricing and per‑unit build cost. Base case
assumes ~410k GPUs, 75% utilization and $8.5/hr rent. If rental rates run
$7–$10/hr, GPU‑leasing ROIC would vary about 23%–39%.
