The AI Cycle - Where We Are Now And Where We're Going

Jim Bianco examines whether the AI investment boom in data centers and computing infrastructure is still in its early stages or beginning to resemble a market bubble. The discussion focuses on the implications of heavy AI-related capital spending for financial markets and the broader economy, with no specific financial results, forecasts, or market moves disclosed.

Analysis

The relevant question is not whether AI demand persists, but whether incremental capex continues to earn returns above the cost of capital. The most crowded exposure remains hyperscaler-led compute, where valuation assumes both sustained utilization and a rapid conversion of inference demand into high-margin software revenue. A modest deceleration in cloud growth or a 200-300bp miss on AI-related gross-margin expectations would likely compress multiples first in NVDA, AVGO and data-center REITs such as EQIX/DLR, even if absolute spending remains elevated.

The less-appreciated bottleneck is power delivery rather than chips. Utilities and electrical-equipment suppliers with booked transmission, switchgear and grid-modernization demand—ETN, PWR, GEV and VRT—have more durable revenue visibility, but their valuations increasingly embed execution perfection; transformer availability, interconnection queues and local permitting can defer data-center revenue by quarters. Over 6-18 months, constrained power availability may shift bargaining power from cloud operators to regulated utilities and independent power producers, though regulatory lag limits near-term earnings capture.

Consensus treats AI infrastructure capex as additive. The contrarian case is substitution: enterprise IT budgets, buybacks and non-AI cloud workloads may be displaced rather than supplemented, leaving aggregate technology spending less resilient than headline capex suggests. The immediate tradable catalyst is quarterly disclosure of capex, depreciation, cloud growth and AI monetization by MSFT, AMZN, GOOGL and META; the thesis is falsified if inference revenue and cloud margins accelerate simultaneously despite rising depreciation.

Key Decisions for Investors

• Avoid adding directional beta to NVDA/AVGO into the next hyperscaler reporting cycle absent evidence that AI revenue is outpacing depreciation expense; use a 5-7% post-earnings drawdown as a better entry point rather than chasing sentiment-driven strength.
• Maintain a 3-6 month relative-value basket: long ETN and PWR versus short an equal-dollar basket of DLR and EQIX. The trade expresses power-grid scarcity over data-center capacity expansion; exit if utility interconnection timelines improve materially or data-center leasing spreads reaccelerate.
• For a defined-risk hedge on crowded AI beta, buy 3-6 month QQQ put spreads financed only partially with upside call overwrites, sized as portfolio insurance rather than a standalone short. The catalyst is a capex-guidance reset or cloud-margin miss; invalidate the hedge if aggregate hyperscaler capex guidance rises while operating-margin guidance is maintained.
• Monitor MSFT, AMZN, GOOGL and META for the ratio of incremental capex to incremental operating income over the next two quarters. A sustained deterioration is a signal to rotate from semiconductors and data-center REITs toward cash-generative software and grid infrastructure; no broad AI short is warranted without that confirmation.