AI Capex Overbuild

Hyperscaler spending on AI infrastructure keeps rising. Seven builders spent $657bn over the four quarters to Q2 2026, with full-year 2026 projections above $720bn, and every major builder raised or held guidance in the last earnings season. The pressure has shifted from the size of the spend to its financing: long-end yields are at multi-decade highs in both the US and UK, bubble warnings from Dalio, Panmure and multiple analysts have accumulated without an offsetting capex cut, and mainstream financial coverage is asking whether AI debt issuance is crowding other borrowers out of the bond market. A projected $1.5tn in AI-linked debt issuance faces tightening conditions. Rate sensitivity has spread beyond hyperscalers into the power infrastructure layer, with the COHR-ANET divergence now the sharpest intra-stack signal of rate-driven disaggregation. Oracle, rated BBB- by S&P, remains the outlier on leverage, with free cash flow of -$5.4bn last quarter. Q3 earnings from late October are the next concrete test. NthEffect tracks the cascade a financing or returns shock would trigger, from GPU and memory suppliers through power and cooling, data-centre real estate and AI-linked credit.