AI industry not in a bubble, but stocks could see correction, SK chief says

SK Group chairman Chey Tae-won said the AI industry is not in a bubble but warned AI stocks have risen "too fast and too much" and could face corrections as valuations overshoot fundamentals. SK Hynix, a key supplier of high-end memory for Nvidia-powered AI data centers, has surged 214% over the past year, reported a record quarterly profit in October and says its chip production for next year is fully booked as it expects an extended chip "super cycle" amid trillions of dollars invested in AI data centers.

Analysis

Market structure: The immediate winners are memory suppliers (e.g., SK Hynix 000660.KS) and GPU leader NVDA, plus hyperscalers (AMZN, GOOGL, MSFT) buying capacity; losers are highly valued AI application/software names that rely on multiple expansion rather than cash flow. SK Hynix’s “sold out” guidance implies pricing power in DRAM/HBM for ~12–18 months, supporting chipmakers’ margins even if AI equities mean-revert. Cross‑asset: a material correction in AI equities would likely spike equity volatility, bid U.S. Treasuries (yields ↓) and strengthen safe‑haven FX; commodity demand (power, copper) for data centers supports energy/utilities for quarters.

Risk assessment: Tail risks include new export controls (China/Taiwan/US) or rapid software efficiency gains that reduce HBM demand, any of which could cut revenues by 20–40% for memory makers within 12–24 months. Short term (days–weeks): profit taking and higher IV around earnings; medium (3–12 months): capex cadence and DRAM price indices will drive realized revenue; long term (2–4 years): overcapacity risk if suppliers ramp production, potentially reversing today’s premiums. Hidden dependencies: hyperscaler procurement timing and Nvidia architecture changes (HBM requirements) are single‑point risks.

Trade implications: Tactical longs in hardware (SK Hynix, NVDA) should be size‑controlled and hedged; expect using 3–6 month protective puts or call spreads during earnings windows. Relative trades: long memory/manufacturing exposure versus short high‑multiple AI software names or thematic ETFs to capture mean reversion. Rotate 3–5% from long-duration growth into data‑center capex beneficiaries (semicap equipment, power infra, DLR) over the next 1–3 months.

Contrarian angle: Consensus fear of an AI “bubble” underestimates near-term memory scarcity — price premium may persist 12–18 months, so buying selective hardware on disciplined pullbacks (15–25%) is justified. Conversely, the market may be underestimating a 24–36 month oversupply shock; therefore avoid one‑way levered bets and use option hedges or pair trades to capture asymmetric outcomes.

Key Decisions for Investors

• Establish a 2–3% long position in SK Hynix (000660.KS) on a pullback of ≥15% from current levels; target +30–50% over 6–12 months if DRAM/HBM pricing stays strong, place a 12% stop-loss or buy a 3–6 month 10% OTM put to hedge.
• Maintain/establish a 1–2% core long in Nvidia (NVDA) for 3–12 months but hedge earnings risk by buying a 3‑month 8–12% OTM put (cost‑efficient long put or put‑spread); trim position if NVDA rallies >30% from entry.
• Implement a relative‑value pair: long 3% notional SK Hynix (hardware exposure) funded by a 2–3% short in high‑multiple AI application names or thematic ETF (e.g., ARKK) to capture mean reversion over 3–9 months.
• Buy a 1–2% notional 3‑month put spread on an AI basket (e.g., NVDA + top software names) to protect portfolio IV spikes around NVDA and SK Hynix earnings; structure as buy 10% OTM put / sell 20% OTM put to limit cost.
• Reduce long‑duration growth exposure by 3–5% over the next 30 days and reallocate into data‑center beneficiaries (2–3% into DLR or leading semicap equipment stocks) to capture infrastructure upside from AI capex.