Google's AI capital expenditure is driven by the existential threat AI poses to its core Search business, rather than traditional blended ROI metrics, creating a "prisoner's dilemma" for continued investment in Gemini. While internal compute returns are lackluster due to struggles in agentic coding and Gemini monetization, external compute sold via Google Cloud to entities like Anthropic and OpenAI yields strong returns due to supply constraints. The sustainability of the AI trade hinges on model developers' ability to grow revenue and profitability fast enough to absorb increasing compute, which appears to be happening with rapid ARR compounding and margin improvements at Anthropic and OpenAI.
Hyperscaler AI compute spending is structurally locked by competitive threats to core businesses, ensuring sustained demand for compute capacity regardless of blended ROI.