'The bubble is ahead of us': hedge fund exec says investors still don't get how big AI is

Bridgewater co-CIO Greg Jensen warns that the real AI-driven bubble is still ahead as capital spending and competition accelerate, entering a "dangerous" resource-grab phase that is already creating bottlenecks in datacenter land, advanced chips and talent (he estimates fewer than 1,000 cutting‑edge AI scientists globally). Jensen argues that AI investment is large enough to move macro indicators — roughly one percentage point of US GDP growth this year — and that headline winners are masking broader market underperformance, leaving many investors unprepared for intensified competition, geopolitical reshaping and further capital flows into AI.

Analysis

Market structure: The immediate winners are AI infrastructure and enablers—GPU leader NVDA, cloud providers (MSFT, AMZN, GOOGL), fabs (TSM, ASML), data‑center REITs (EQIX, CONE) and commodity/mining names supplying copper/energy; losers are legacy software/services, small‑cap domestically oriented firms, and labor‑intensive incumbents. Capital is shifting from broad market breadth into concentrated megacaps and real assets (land, power, chips), increasing concentration risk and making pricing power stickier for enablers; expect top‑5 tech to outperform broad US equities by a further 5–15% over 6–12 months if Jensen is right.

Risk assessment: Key tails include (1) hard regulatory clampdowns or export controls that remove Chinese demand (6–18 months), (2) chip‑supply shock or ASML/TSM outages causing 20–40% GPU price swings, and (3) an AI “accident” triggering immediate de‑risking across tech. Short horizon (days–weeks) is dominated by sentiment/flow volatility; medium (3–12 months) by capex guidance and supply constraints; long (1–3 years) by productivity gains and consolidation. Hidden dependencies: <1,000 top researchers, power grid limits and geopolitically concentrated tool suppliers create single‑point failures.

Trade implications: Favor concentrated exposure to NVDA (direct) and data‑center REITs (EQIX/CONE) while hedging market beta; buy 6–12 month call exposure rather than naked equities to limit downside and capture convexity. Rotate out of high‑valuation non‑AI growth names and small caps (IWM) and allocate 1–2% to copper/materials (COPX/FCX) and energy utilities to hedge rising power demand.

Contrarian angles: The consensus underestimates resource scarcity and real‑economy capex: AI spending is already ~1ppt GDP impact this year per Jensen — that suggests cyclical commodity reflation, not just a software bubble. Mispricings: data‑center landlords and utilities are underowned relative to NVDA/MSFT; historical dot‑com parallels overstate speculative retail froth—this cycle is CAPEX‑intensive with longer lasting real returns. Watch for unintended consequences: grid strain, talent bidding wars inflating margins but slowing R&D.

Key Decisions for Investors

• Establish a 2–3% tactical long in NVDA (ticker NVDA) via 9–12 month ATM calls (or 100% equity equivalent) to capture further GPU scarcity-driven upside; trim half if NVDA rallies >20% from entry or implied vol rises >30 pts.
• Add 1–2% combined exposure to data‑center REITs: 0.75–1% EQIX and 0.25–0.5% CONE, hold 12–24 months; set stop/exit if same‑store NOI growth falls below +5% YoY across two consecutive quarters.
• Implement a pair trade: long NVDA (2%) funded by a 1.5% short position in IWM (Russell 2000 ETF) to express concentration outperformance while hedging small‑cap downside over 3–9 months.
• Allocate 0.5–1% to copper/industrial metal exposure via COPX or FCX for 6–18 months to front‑run increased energy/chip metal demand; take profits if copper spot rises >25% or if fab capex cancellations are announced.
• Purchase tail protection: buy 3‑month SPY puts 5% OTM sized to cost ~0.5–1% of portfolio or VIX 1–2 month calls to hedge against regulatory/AI‑accident shock events expected within the next 90 days.