Will capitalism end capitalism?

Key datapoint: Anthropic’s CEO warned AI could eliminate ~50% of entry-level white-collar jobs within five years. The author argues that as digital labour approaches the marginal cost of electricity, wages—and thus mass consumer demand—could collapse, hollowing out sectors and regions and concentrating productive power in a few AI owners (notably US and China). Policy shortfalls on redistribution, competition and portable social security raise the risk of severe social and economic disruption during the transition.

Analysis

Concentration of AI productive capacity creates a two-way pressure: outsized profit capture for infrastructure owners and accelerating political/regulatory scrutiny that can crystallize quickly. Expect the lion’s share of incremental AI spend to flow to a handful of compute, data‑center and foundry providers over the next 24–36 months, widening their free cash flow margins while making them primary targets for break-up, profit‑taxes or operational constraints.

Energy and real‑estate arbitrage will be a second‑order force: datacenter builds will follow lowest‑cost, dispatchable power and favorable permitting, shifting capex and tax bases to inland US regions and select international hubs. That creates asymmetric winners among utilities, transmission owners and local munis (near‑term beneficiaries of hosting fees) while hollowing out regions with high electricity costs or weak grid capacity.

Supply‑chain choke points—advanced logic wafers, specialty DRAM/HBM and custom networking—translate directly into pricing power for select suppliers for at least 18–36 months; tactical shortages and lead‑time dynamics will sustain margins through a multi‑year capex cycle. The main catalysts that could derail this trajectory are rapid, coordinated regulatory action (months–2 years), export controls that fragment supply chains, or a political push to redistribute AI rents (taxes/subsidies) which would compress net returns.

Contrarian lens: the market underweights incumbent incentives to monetize AI via services and price discrimination rather than pure wage displacement, which preserves demand into durable goods and premium services. Trade accordingly: own the infrastructure, hedge the socioeconomic demand risk, and size for policy shock volatility rather than pure technology failure.

Key Decisions for Investors

• Long NVDA (or NVDA 2027 LEAP calls) — 12–36 month horizon to capture continued AI compute demand. R/R: asymmetric upside (3–6x) if model training spend continues; downside limited to premium but keep position size <4% NAV due to regulatory risk.
• Pair trade: Long EQIX (data‑center REIT) vs short KRE (regional banking ETF) — 6–24 months. Rationale: hosting fee growth and sticky contracts lift REIT cash flow while payroll/deposit stress and regional exposure to employment cycles pressure banks. Target +25–40% vs -20–30%; hedge with REIT put if banks rally.
• Long TSM (foundry exposure) or a basket of leading fabs (TSM, ASML via indirect plays) — 18–36 months. Semiconductor supply tightness sustains pricing; payoff is 30–60% on sustained AI capex. Risk: demand re‑profile or export controls shunting business away.
• Tail hedge: Buy 3–12 month VIX calls or a modest allocation to long‑dated sovereign bonds (TLT) — protect portfolio against regulatory/political shock and market‑wide risk‑off. Cost: small drag on carry but protects concentrated tech/infrastructure exposure.