Get used to the K-shaped economy. It’s likely here until 2035, thanks to AI’s outsized benefit for the wealthy

Oxford Economics estimates AI investment has delivered a greater-than-7% uplift in U.S. household wealth, but the gains are concentrated among high-income households, reinforcing a K-shaped economy that McFee says is likely to persist through at least 2035. Federal Reserve distribution data cited show the bottom 50% held $4.25tn by Q3 2025 (up 1,189% since 2010) while the top 0.1% held $24.89tn (up 281%), illustrating growing polarization. Oxford modelling and Penn Wharton analysis point to an S-curve in AI adoption with a plateau in the early 2030s and a likely hollowing-out of middle-skilled roles, implying stronger consumption and asset returns for the wealthy but continued labor-market pressure and uneven productivity gains for lower-skilled workers.

Analysis

Market structure: The AI capex wave concentrates rents to platform/cloud (NVDA, MSFT, GOOGL, AMZN), semiconductor fabs/equipment (SOXX, ASML, LRCX) and data‑center real estate (EQIX); these see pricing power and 5–15% EPS tailwinds over 12–24 months. Middle‑skill service providers and task‑based SaaS face structural demand erosion as adoption follows an S‑curve (rapid adoption 2026–2032) producing a hollowed middle and stronger demand for power, copper and specialty silicon.

Risk assessment: Key tail risks are export controls/antitrust actions vs. large AI suppliers, a capex bust if enterprise ROI disappoints, and energy/grid constraints raising operating costs for data centers. Near term (days–months) watch earnings/capex guidance and Fed comments; medium (6–18 months) watch corporate capex cadence and jobless claims; long term (3–10 years) the distributional effects and regulation will shape revenues and consumer demand.

Trade implications: Favor concentrated long exposure to AI leaders and semicap OEMs (2–3% NVDA, 2% MSFT, 3% SOXX) plus data‑center REITs and industrials benefiting from site builds (EQIX, HD, FAST). Hedge with shorts or underweights to consumer‑retail ETF exposure (XRT) and staffing/middle‑skill service names; implement 3–6 month call spreads on NVDA to express upside while capping cost and sell OTM calls on cyclicals to fund hedges.

Contrarian angles: Consensus underestimates hardware/infra winners and trade contractors (plumbers/electricians) needed for data centers — these are early, low‑visibility beneficiaries that historically re-rate late in cycles (analogue: 1990s internet capex). The market may be underpricing regulatory tail risk and a possible mid‑cycle capex pause; alternatively, faster low‑skill augmentation could compress the K‑shape sooner, creating mean reversion opportunities in mid‑skill exposed equities.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA within 30 days; prefer a 3–6 month call‑spread (buy calls ~+20–30% strike, sell higher strike) to limit premium if IV >30%.
• Allocate 3% to SOXX (semiconductor ETF) and 1.5% to EQIX (data‑center REIT) for 12–24 months to capture hardware and real‑estate tailwinds; add on any pullback >10% from entry price.
• Initiate a relative trade: long SOXX (3%) vs short XRT (retail ETF) (2%) — expect tech/infra outperformance over consumer retail over next 6–12 months as wealth effect concentrates at top 10%.
• Add a 1–2% tactical long in MCO (Moody's, ticker MCO) over 12–24 months — ratings/structured product activity and complexity from AI financing should lift fees; trim if downside to credit spreads >50bps widens.
• Reduce exposure to mid‑skill dependent staffing and low‑end retail by ~50% within 90 days; redeploy proceeds into industrials (HD, FAST) and semicap names. Monitor three catalysts to re‑rate: quarterly capex guidance, major AI regulatory announcements, and US non‑farm payrolls — act within 7 trading days of material changes.