Bernstein projects sustained hyperscaler CapEx and $1.3T enterprise AI spending despite potential scaling law challenges

Bernstein analysts suggest that even if AI scaling laws face issues, hyperscaler capital expenditure (CapEx) for AI infrastructure is expected to remain robust for several quarters. They estimate a $1.3 trillion enterprise willingness to pay for AI, driven by continuous model improvements that unlock new use cases and boost productivity. This ongoing improvement drives demand for both training and inference, benefiting AI infrastructure, including servers and memory. Bernstein identifies model improvement as a better leading indicator than hyperscaler CapEx, which they view as a lagging indicator. They anticipate healthy near-to-medium term data points due to consistent model advancements. The report also suggests that in a potential digestion cycle, there would be ample time to adjust investments, as any breakdown in scaling laws would likely take a year to become evident and hyperscaler CapEx would continue for several quarters thereafter. The intelligence revolution is seen as structurally positive for AI server and memory vendors.