By Djellal Djouad
Position taken. The CrossVol desk does not equivocate. The US AI infrastructure complex is in a bubble by any honest metric — capex to revenue, debt maturity to depreciation, optionality to fundamentals.
Almost every sell-side desk has converged on the same narrative: hyperscaler capex grows 25–30% CAGR through 2028, US power demand doubles by 2027, memory stocks enter a supercycle extending through 2028. The data tables across Goldman, Morgan Stanley, JPMorgan are nearly identical. NVDA, Vertiv, the datacenter REITs, the power complex — all are priced for the consensus.
The consensus is wrong, and it is wrong on quantifiable grounds.
Five arguments — none speculative, each resolves on a specific date or against a specific number:
Hyperscaler capex is outrunning hyperscaler revenue — 165% capex expansion against 47% AI/cloud revenue expansion over 2023–2025. By 2028, $1.4T of AI infrastructure to depreciate against revenue streams not yet written.
The corporate bond wall — $121B long-dated IG debt issued by hyperscalers in 2025 (4.3× prior-decade average), tracking $230–240B in 2026. Pension funds and insurers absorb the issuance on spread alone. The repricing happens in days when revenue assumptions get revisited.
Token commoditization is already underway — DeepSeek V4 Pro at $0.87 per million output tokens vs. $25–30 for US frontier. Confirmed permanent on 22 May 2026. This is the exact pattern of every commoditization cycle in tech history (storage 2007, solar 2011, LED 2014).
The US power grid bottleneck is math, not narrative — PJM 2026/2027 capacity cleared at $329.17/MW-day (+11.4× in two years), ~50% of planned data center projects delayed or cancelled. Transformers take 18 months to manufacture, high-voltage corridors 5 years to permit.
The dot-com analog is tighter than everyone wants it to be — capex front-loading against unproven unit economics, financed by long-dated debt, sold to retail and pension allocators on a single dominant narrative. Cisco, Lucent, JDS Uniphase fell 80–95% between March 2000 and October 2002.
The window: Q4 2026 → Q1 2027.
Two dated catalysts:
10 November 2026 — US-China tariff truce expires
27 November 2026 — Chinese gallium/germanium/antimony export controls suspension expires
The CrossVol desk expects pure AI-infrastructure equities to draw down 25–40% across that window.
Read the full piece in 11 languages: crossvol.com/en/blog/us-ai-bubble-2026
The complete framework, with trade construction: The China AI Disruption Thesis — 14 chapters, dated catalysts, dispersion / variance / single-name shorts / credit hedges. Read on Amazon Kindle · Book page
The four-lens market-reading framework that produced the call: Beyond Gamma Exposure — Amazon Kindle · Book page
CrossVol Terminal — crossvol.com
CrossVol Team & Djellal Djouad
Further reading