Goldman's framework on hyperscaler AI revenue versus capex is worth picking apart. They set the breakeven bar at $308B annual AI revenue based on average 2026-2027 capex, with higher ROIC targets pushing that requirement to $417B, $526B, and $636B at 10%, 15%, and 20% respectively.

The issue is their baseline treats GenAI spend as an isolated product line rather than a platform-wide re-acceleration driver across core cloud services. Cloud revenue growth for $Amazon.com(AMZN)$  $Alphabet(GOOGL)$  $Microsoft(MSFT)$  and $Oracle(ORCL)$  accelerated from +25% in 2024 to +48% in 2Q26. Current AI/cloud revenue is running roughly $70B annualized above the pre-AI trend. Matching incremental capex against a static run-rate ignores what happens when growth accelerates by +23 percentage points over a two-year span.

This capex also isn't speculative "build it and they will come" infrastructure. It's backed by committed commercial demand, with announced backlogs above $1.5T to $1.7T.

Goldman's breakeven math front-loads capital recovery into a compressed window. A standard 5-to-7-year straight-line depreciation schedule would require much lower annual revenue run-rates in the early years to still generate positive IRR over the lifetime of the infrastructure.

Hyperscalers are building capacity to lock in enterprise migrations for the next decade and prevent churn to rival platforms. Even if GenAI were removed entirely from the picture, hyperscale data center capacity would still need to expand quickly. The core IT transformation that predates the GenAI boom continues to outstrip legacy data center supply. AI just accelerated a trend that was already compounding.