An OpenAI cofounder ‘vibe coded’ an analysis of the U.S. labor market’s exposure to AI, and the highest-paying jobs have the worst scores

Karpathy's quick analysis put the overall AI exposure score at 4.9, with occupations earning >$100k averaging 6.7 vs < $35k averaging 3.4; many white-collar roles (software developers, data scientists, financial analysts, writers) scored 9 while manual roles (construction laborers, roofers, janitors) scored 1 and care/service roles scored 2. The chart was removed after pushback; Anthropic's report finds actual AI adoption lags feasible capability, and Citadel counters doomsday claims noting Indeed software-engineer postings are up ~11% YoY in 2026 and little evidence yet of widespread displacement. Implication: theoretical disruption risk is concentrated in higher-paid, knowledge-work roles, but near-term labor-market signals and demand for compute/infrastructure are moderating immediate market disruption.

Analysis

AI-driven substitution is creating a multi-year capital cycle that is underpriced: hyperscalers and GPU vendors capture disproportionate returns because they internalize scale, spare parts, and power-efficiency gains. Expect a stretch of elevated capex for datacenters, specialty cooling and high-voltage power upgrades over the next 12–36 months, which compresses margins for downstream software/service providers that cannot monetize raw compute.

On labor, the immediate arb is not pure job loss but a supply-side squeeze: fewer traditional entry-level roles mean slower skill on‑ramps, which increases scarcity premia for mid-to-senior engineers and ops staff. That dynamic will raise labor cost for firms dependent on experienced talent while lowering churn and lifetime hiring budgets for firms that successfully industrialize AI workflows.

Catalysts that will re-rate these thematic winners/losers are measurable and relatively short-dated: quarterly capex announcements from top cloud providers, quarterly GPU shipment data, and firm-level guidance on hiring mix. Reversals are plausible if marginal compute prices spike (raising substitution costs), meaningful regulatory constraints on model deployment emerge, or integration/time-to-value proves longer than vendors forecast — any of which could materially slow the capex cycle and reflate human labor demand.

Key Decisions for Investors

• Long NVDA (NVDA) — buy 12-month LEAP calls sized 3–5% of portfolio. Rationale: convex exposure to GPU-driven compute TAM; upside 2–4x if hyperscaler capex stays elevated, downside limited to premium paid (~50% downside if adoption stalls). Entry: within next 2–6 weeks ahead of expected vendor earnings; exit or trim on >80% realized move or if guidance weakens.
• Long Equinix (EQIX) or AMZN (AMZN) cloud exposure — buy EQIX shares or a 6–18 month call on AMZN sized 2–4% notional. Rationale: direct beneficiary of increased colocation and interconnect demand; reward realized through lease-rollup and pricing power. Risk: slower-than-expected enterprise cloud migration; stop-loss at 15% drawdown or if hyperscaler capex guidance falls quarter-over-quarter.
• Long cybersecurity names (CRWD or PANW) — buy 9–12 month calls or 3–5% long equity position. Rationale: AI increases attack surface and compliance spend, creating durable SaaS upsell opportunities. Risk/reward: expect 30–100% upside if enterprise adoption accelerates; downside capped by subscription stickiness.
• Short staffing/external-entry-level placement (MAN) — initiate a small 1–2% short position or buy downside put spread for 6–12 months. Rationale: secular decline in entry-level openings compresses placement volumes and fee churn. Risk: temporary rebound in hiring or a pivot to higher-margin staffing services — cap position size and hedge with a market-neutral overlay.