An ’AI winter’ is likely within one to three years, BCA warns

BCA Research warns an "AI winter" is increasingly likely within one-to-three years, arguing that AI-related CAPEX and data-centre construction may slow and trigger a significant repricing of tech/growth stocks. Its base case (80% probability) sees only moderate macro productivity gains of roughly 0.4–0.5% annually, with just a 5% chance of AGI and a 15% downside scenario tied to data-centre misallocation; BCA estimates $9–$12 trillion of recent market gains are unexplained by earnings or rates. The note highlights rising debt issuance by hyperscalers, risk of a compute glut, and recommends underweighting tech enablers while favoring adopters (financials, pharmaceuticals, biotech, life sciences, defense) until valuation excesses are worked off.

Analysis

Market structure: The likely “AI winter” scenario shifts winners toward AI adopters (financials, pharma, defense, life sciences) that extract productivity gains without heavy CAPEX, and away from enablers (data-center builders, chip-capex-heavy suppliers). Expect pricing power compression and lower ROIC for enablers if CAPEX overshoots demand; a 10–30% mid-cycle revenue/EBIT multiple reset for highly valued enablers is plausible within 12–36 months. Credit issuance by hyperscalers raises supply of corporates and can push spreads +50–150bps in weaker BBB/BB issuers if sentiment shifts.

Risk assessment: Tail risks include a severe compute-glut writedown (10–40% capex impairment across select suppliers), abrupt export controls/regulatory constraints on large models, or an energy-price spike that raises operating costs for data centers by >15%. Near-term (days–weeks) moves will be sentiment-driven around earnings/CAPEX guide; medium-term (3–12 months) depends on bond issuance and adoption metrics; long-term (2–5 years) on realized productivity gains (BCA expects +0.4–0.5% GDP uplift). Hidden dependencies: power grid constraints, chip lead times, and corporate procurement cycles can amplify or delay the downturn.

Trade implications: Favor long exposure to XLF and selective healthcare/defense (IBB, LHX) as 6–18 month plays; underweight or hedge enablers (SMCI) with options because valuations price a strong AI tail. Use pair trades (long financials/biotech, short infra-enablers) and buy put protection on indices and 3–9 month puts on top enabler names. Monitor bond issuance calendars and hyperscaler CAPEX guidance as trade triggers.

Contrarian angles: Consensus underestimates the sticking power of vertical AI adoption—some enablers with asset-light hybrid models may be mispriced; SMCI and AppLovin pockets of idiosyncratic strength could persist. The reaction could be overdone in credit: sell-off in BBB paper may create selective buying opportunities. Historical parallels (telecom overbuild 2000s) show severe losses for a few suppliers but durable cash flows for diversified adopters.

Key Decisions for Investors

• Establish a 2–3% long position in XLF (Financial Select Sector SPDR) over 6–12 months, target +15–25% upside if AI benefits rotate to financials; set a 10% trailing stop to limit drawdown.
• Initiate a dollar-neutral pair: short SMCI (Super Micro Computer) equal to 1–2% of portfolio notional and go long IBB (iShares Nasdaq Biotechnology ETF) 1–2%; hold 3–12 months and rebalance on earnings/CAPEX guides.
• Buy 3–6 month put spreads on top enablers (e.g., SMCI) sized to 1% portfolio risk — buy 10% OTM puts and sell 20% OTM puts — to cost-effectively express downside if CAPEX demand falls sharply.
• Trim high-multiple cloud/data-center infrastructure exposure by ~30% within the next 4–8 weeks and redeploy proceeds into healthcare/defense ETFs (IBB, LHX) and a 1% allocation to S&P 500 3-month 5% OTM puts as portfolio insurance against a tech-driven drawdown.