‘Yes, AI Is a Bubble. There Is No Question.’

Private-sector AI spending is forecast to exceed $700 billion in 2026, while recent agent-driven revenue surges — Anthropic doubling revenue in two months and OpenAI adding roughly $1B of annualized revenue per week — have complicated the bubble thesis. Investor Paul Kedrosky warns AI is likely one of the five largest CapEx bubbles in history, potentially representing a material share of GDP growth (he cites 50–80% in some measures) and risking rotating crashes as overbuilt assets fail to service debt. Stripe and other data points show unusually rapid revenue monetization, creating a tension between historic capex intensity and current sales that raises sector-wide downside risk, private-credit stress, and consolidation pressure among hyperscalers. Portfolio action: reduce concentrated exposure to capex-heavy/lease-dependent names, monitor private-credit and hyperscaler counterparty risk, and favor businesses showing durable, diversified revenue capture from AI adoption.

Analysis

The most durable non-obvious bifurcation from the current thematic is between “compute stack” beneficiaries (chips, fab equipment, hyperscaler-owned GPUs) and the capital-heavy real-estate and financing layer (colocation, leased data-centers, private lenders). If hyperscalers monetize agent-driven software directly (high GP margin SaaS-like economics), they will capture margin and leave colo landlords and non-core hosting providers with stranded capacity and long-duration lease mismatches; vacancy/rent pressure could compress REIT yields by 200–400bps over 6–18 months. Meanwhile, chipmakers with architecture-driven moats can re-price profits into higher ASPs for a multi-year cycle, insulating them from a real-estate correction even if overall AI capex growth slows.

Credit is the most underpriced channel for a messy unwind. A capex-heavy buildout financed by nonbank lenders and covenant-lite structures creates a high sensitivity of mid-tier creditors to a 30–50bp move in U.S. policy rates or a 3–6 month growth pause in cloud procurement. That path produces concentrated defaults in niche owners/operators (small colo, GPU-leasing startups) well before headline tech writedowns show up in large-cap indices; recovery values will depend on repurposability of specialized cooling/power rigs.

Catalysts to watch: hyperscaler quarterly guide-backs (1–3 months) and incremental pricing on instance-level GPU offerings (3–6 months) that reveal demand durability; bank and private-credit lender earnings and utilization metrics (3–9 months) that will signal stress migration; and 6–24 month capex cadence from the top three cloud providers which sets the speed of any consolidation. The contrarian angle: if revenue gains from agents are real and sticky to enterprise ARR, overbuilt infrastructure could simply consolidate into fewer owners and fund a multi-year hardware upcycle — a bullish outcome for high-quality chip-equipment franchises but a structural negative for levered real-asset owners and credit-intermediaries.

Key Decisions for Investors

• Pair trade — Long NVDA (or buy NVDA 12–18 month calls) / Short EQIX or DLR (buy 9–12 month put spreads): Thesis: NVDA captures secular GPU ASP upside; colocation REITs trade to reflect lease overhang. Target R/R: 2:1 — expect 30–50% upside on NVDA calls vs 25–40% downside on REIT puts if rent roll re-prices over 9–12 months.
• Short high-valuation software beta — Short IGV or buy 6–12 month put spread on a concentrated SaaS basket while going long MSFT or AZURE exposure (calls 12 months): Thesis: agent adoption creates an early revenue sugar high then ARR reversion; pick defensible cloud platforms as hedges. Target R/R: 1.5–2:1 if SaaS multiples compress 20–35% within 6–12 months.
• Credit hedge — Buy 6–12 month HYG put spread or purchase protection via single-name CDS on smaller cloud/colo lenders where available: Thesis: private-credit funded capex is the channel for cascading defaults; a modest stress scenario (1–2% rise in HY default expectations) will re-rate HYG. Target R/R: convex payoff with limited premium spend; expect hedge to pay off materially in a 6–18 month tightening/stress window.
• Long semicap / moat plays — Buy ASML or AMAT 12–24 month calls: Thesis: equipment vendors with unique EUV/advanced-node exposure are the clearing value in a consolidation outcome and should see multi-year pricing power even if infrastructure ownership consolidates. Target R/R: 1.5–3:1 on a multi-quarter to multi-year upcycle driven by restocking and new node investment.