AI Infrastructure

Investment Thesis

Hyperscaler capital expenditures on GPU clusters, data centers, and high-bandwidth memory have entered a multi-year buildout cycle, with major cloud providers committing hundreds of billions in capex through at least 2027. Enterprise AI adoption remains early relative to the infrastructure it will eventually require. Demand for specialized compute, custom ASICs, advanced cooling, and power infrastructure will compound as model scale and inference workloads continue to grow.

Theme Score Breakdown

AI Infrastructure is the only theme in this library that scores Elite because no other trend combines measurable hundred-billion-dollar spending commitments, genuine competitive urgency across multiple hyperscalers, and structural compounding across every layer of the stack simultaneously. Durability drops slightly from the prior score because the specific buildout cycle is concentrated rather than globally distributed — this is a U.S.-and-allied-nation story, not a universal one. Export control constraints are a real ceiling on the addressable market. The -2 risk penalty is modest because the downside scenario (AI ROI disappoints) would slow growth, not reverse it — the infrastructure already being built cannot be unbuilt.

Key Risks

• Demand cycle reversal if enterprise AI ROI disappoints at scale
• U.S. export controls limiting addressable market for chip companies
• Competitive GPU alternatives eroding NVDA pricing power
• Power capacity constraints slowing data center expansion timelines