5 big analyst AI moves: Top picks for 2026 unveiled as AI winter risk grows

Major brokerages raised targets and ratings on key semiconductor and AI-related names amid tightening supply and strong AI demand: Morgan Stanley lifted Nvidia's price target to $250 and Broadcom to $443 citing constrained data‑center availability through 2026 and stronger-than-expected Google TPU builds; BofA raised ASML's target to €1,158 expecting lithography intensity and FCF to drive a multi-year upswing; TD Cowen put AMD on its Best Ideas 2026 with a $290 target, and Summit upgraded Marvell to Buy, forecasting data‑center growth of ~25% in FY27 and ~40% in FY28 with January-quarter revenue of $2.20bn (+6% seq.) and data‑center +9%. Offsetting this optimism, BCA Research warns of a plausible 'AI winter' within 1–3 years, assigning only a 5% chance of AGI and an 80% base case of modest productivity gains (~0.4–0.5% p.a.), arguing that markets may have priced in $9–$12 trillion of gains since late 2022—advising caution on tech valuations and positioning.

Analysis

Market structure: Winners are chipset and lithography leaders (NVDA, AVGO, ASML, AMD) as constrained front-end (3–5nm) and back-end (CoWoS, HBM) capacity gives incumbents pricing power and order visibility through 2026–27. Losers include undifferentiated foundries/IDMs (INTC) and smaller AI-ODM suppliers who lack secured wafer/assembly slots; memory makers face a near-term “gold rush” (BofA sees lithography intensity to ~26% by 2028) that could drive volatile cycles.

Risk assessment: BCA’s framework (5% AGI / 80% moderate productivity / 15% adverse) implies high valuation tail risk — a 20–40% drawdown in hyperscaled capex would catalyze an “AI winter.” Immediate (days–weeks): reorder and supply updates; short-term (3–12 months): demand cadence from hyperscalers and TPU ramps; long-term (2026–28): ASML-driven re-rating or memory oversupply. Hidden dependency: hyperscalers’ bond-funded capex; increased leverage raises systemic liquidity risk if returns disappoint.

Trade implications: Tactical long bias to NVDA/AVGO/ASML with explicit hedges — favor 6–12 month call spreads (to capture supply-driven upside) and pairs (long AVGO / short INTC) to isolate AI-specific upside vs legacy CPU risk. Options: buy 3–6 month QQQ puts (10% strike) as tail insurance. Rotate modestly from pure enablers into adopters/defensive cyclicals (financials, pharma) over next 3–9 months.

Contrarian angles: Consensus underestimates timing risk — revenue upside is supply-constrained not demand-driven; Broadcom’s TPU exposure risk (GOOGL homegrown TPU) is real but multi-year, so short-term multiple expansion may continue. Historical parallel: 2017–19 memory cycles show rapid FCF reversals; monitor WFE orders down >15% QoQ or NVDA data-center revenue guidance missing by >10% as hard stop triggers.

Key Decisions for Investors

• Establish a 3% portfolio long in NVDA via a 9-month call spread (buy ATM, sell +20% strike) to capture constrained supply-driven upside; trim if NVDA management revises data-center revenue guidance down >10% on next call.
• Build a 2–3% position in ASML (buy shares or Jan-2027 calls) as a 2026–2028 structural play; reduce by 50% if quarterly WFE orders decline >15% QoQ or China revenue falls below low-20% guidance range.
• Pair trade: go long 1.5% AVGO equity and short 0.75–1% INTC to express Broadcom’s design-in/TPU exposure vs Intel’s lagging node competitiveness; rebalance if AVGO/INTC spread moves >25% from entry.
• Buy a 3-month QQQ 10% OTM put (size = 2% portfolio) as tail hedge against an AI-winter-driven tech drawdown (>15%); consider rolling if implied vol rises >50% from current levels.
• Reduce high-beta small-cap AI-enabler exposure (e.g., SMCI/APP-sized positions) by 20% and reallocate proceeds to select financials (BAC +1–2% overweight) and pharma/biotech exposure over the next 6–9 months to diversify growth-risk.