The Macro
71High ConfidenceTrending Down ↓getthemacro.com

The AI Capex Supercycle Will Send Semiconductor and Power Infrastructure Stocks Soaring

Hyperscaler AI capex is exceeding $250B in 2025 and shows no signs of slowing. This unprecedented infrastructure buildout creates a multi-year demand supercycle for the supply chain that enables it — GPU makers, semiconductor equipment, data center cooling, and power utilities. Enterprise AI adoption is in early innings, with every Fortune 500 company building an AI strategy, and new compute-hungry modalities like video generation and AI agents are creating step-function increases in demand.

Investment Plays

Good For

SMHETFLong

VanEck Semiconductor ETF

Broad semiconductor exposure across the AI supply chain.

NVDAStockLong

NVIDIA Corporation

Dominant AI GPU provider with 80%+ market share. Highest beta play on AI capex.

VRTStockLong

Vertiv Holdings

Data center cooling and power management. Critical infrastructure for AI data centers.

Bad For

INTCStockShort

Intel Corporation

Losing share in AI/data center to NVDA and AMD. Manufacturing restructuring adds uncertainty.

For & Against the Thesis

For

  • AI model capabilities continue scaling with compute — no plateau in sight for frontier models
  • Enterprise AI adoption is in early innings; every Fortune 500 company is building AI strategy
  • Data center construction boom has 3-5 year build cycles, locking in demand regardless of near-term sentiment

Against

  • Enterprise AI ROI remains unproven — spending could pull back sharply if productivity gains disappoint
  • Open-source models and efficiency improvements could dramatically reduce per-task compute requirements
  • Nvidia's margins face compression from AMD, Google TPUs, Amazon Trainium, and custom silicon

Open Questions

Active Questions

(3)

Will enterprise AI ROI materialize enough to sustain current capex levels?

High ImpactUnresolved

If yes — validates the supercycle — spending continues and potentially accelerates | If no — capex cliff as hyperscalers pull back, similar to 2001 telecom bust

Microsoft, Google, Meta all guiding higher capex for 2025; but ROI metrics remain vague

Will efficiency gains reduce compute demand growth?

High ImpactUnresolved

If yes — gPU/data center oversupply emerges, margin compression across the supply chain | If no — jevons paradox — efficiency gains get consumed by new use cases and scale

DeepSeek showed smaller models can be capable, but demand for frontier models keeps growing

Will power constraints become a binding bottleneck?

Medium ImpactPartially Resolved

If yes — slows the buildout but benefits power infrastructure companies (nuclear, utilities, grid) | If no — buildout continues unimpeded; pure infrastructure plays benefit most

Multiple data center projects delayed due to power availability; nuclear PPAs signed by MSFT, GOOG, AMZN

Key Indicators

NVIDIA (NVDA)

Supports ThesisWeight: 1.5x
yahoo·NVDA·Bullish when rising

VanEck Semiconductor ETF

Supports ThesisWeight: 1.2x
yahoo·SMH·Bullish when rising

S&P 500

Supports ThesisWeight: 0.5x
yahoo·^GSPC·Bullish when rising

Utilities Sector (XLU)

Supports ThesisWeight: 0.8x
yahoo·XLU·Bullish when rising