AI
Infrastructure Portfolio Concentration Creates Single-Factor Drawdown
Risk
When >35% of a growth portfolio is allocated to companies whose
primary revenue driver is the same demand source (hyperscaler AI capex),
sector-level diversification is illusory. In a capex cycle reversal,
these names will correct simultaneously regardless of individual
business quality. Two companies can be in different sectors —
semiconductors vs. software — yet both be downstream of the same
hyperscaler spend decision, making them effectively a single bet.
Evidence
- muji portfolio Q1-2026: 37% AI infrastructure concentration (APP
10%, CRDO 10%, NBIS 10%, MU 8%, ALAB 5%, IREN 4%)
- All six names share a common demand source: hyperscaler AI capex.
APP monetises AI-driven mobile advertising; CRDO sells AI datacenter
active electrical cables; NBIS provides AI cloud compute; MU supplies
HBM memory for AI training/inference; ALAB makes AI server connectivity
ASICs; IREN is building AI cloud capacity.
- Analysis explicitly flagged: "This is the correct secular bet for
2025-2027 but introduces concentration risk if AI capex cycles turn."
All six would correct simultaneously in a hyperscaler spend reversal
even though they span ad-tech, connectivity semiconductors, cloud
infrastructure, memory, and ASICs.
- Correction (Bear, April 2026): APP was
mis-classified in the original analysis as an "AI infrastructure
demand-source" name. Bear's portfolio review explicitly rejects this
grouping: "APP is an ad-tech platform that uses AI — its revenue is
driven by mobile app install advertising, not hyperscaler capex
decisions. The demand sources are genuinely different." APP's revenue is
driven by app developer / brand advertiser spend on mobile user
acquisition — a separate budget pool from hyperscaler AI capex. If
hyperscalers cut AI infrastructure spend, APP revenue is not directly at
risk; it is at risk from mobile advertising budget contractions (a
different macro driver). Including APP alongside CRDO/MU/NVDA in a
"hyperscaler AI capex" concentration bucket overstates correlated
drawdown risk.
Implication
When reviewing portfolios with significant AI exposure, apply a
"demand-source concentration" test: sum allocations to companies sharing
the same primary demand driver, not just sector. Flag if any single
demand source exceeds 30% of portfolio. This is a stricter screen than
standard sector concentration — it identifies cases where apparent
diversification (different sectors, different business models) masks a
single-factor dependency on the same underlying spend decision.
Boundary rule (added April 2026): "AI-enabled" ≠ "AI
infrastructure demand-source." A company that uses AI to
improve its product (ad-tech, vertical SaaS, digital health) draws on a
different budget pool than a company that sells to the AI
infrastructure build-out (memory, semiconductors, cables, cloud
compute). Confusing the two inflates measured concentration. The test:
if hyperscalers freeze AI capex tomorrow, does this company's revenue
fall directly? If no — if it depends on brand advertisers, enterprise
software budgets, consumer spending — it does not belong in the AI
infrastructure concentration bucket.