NVIDIA and its GPU customers are now a large driver of equity market returns, earnings growth, earnings revisions, industrial production and capital spending. NVIDIA’s financial results are extraordinary (it beat on revenues and earnings again last week, and announced a $50 billion stock buyback); NVIDIA has also experienced the fastest road to being the market’s largest stock in the post-war era.

But for investors, the more important questions look past the economics of selling GPUs and focus on the ability of hyperscalers (Google, Amazon, Microsoft, Meta, etc) and other AI infrastructure users/providers to earn adequate returns on hundreds of billions in AI-related capital spending. The level of this spending now rivals the mainframe era of the late 1960’s and the fiber optic deployment of the late 1990’s. For adequate returns on AI infrastructure to materialize, within the next 12-18 months we will need to see a greater shift in favor of “inference” tasks (AI used to run production models for corporate customers) rather than GPU capacity primarily being used to train foundational models and chatbots. In this Eye on the Market, we take a closer look.

Watch the Podcast