Citi’s Top Chip Stocks to Buy Now as AI Infrastructure Spend Holds Strong

Citi Research doubles down on four semiconductor names into mid-2026: Broadcom (price target $475, ~40% implied upside) tops the list after a December-quarter-driven consensus EPS revision of +25%; NVIDIA (PT $270, ~52% implied upside) is framed as the cornerstone ahead of GTC and inference/memory architecture debates. Texas Instruments (PT $235, ~25% implied upside) is called out for gross-margin self-help and a ~3% dividend, while Monolithic Power Systems (PT $1,350, ~33% implied upside) is favored for an early-stage power‑semiconductor cycle as AI data centers push power density higher.

Analysis

AI-driven infrastructure demand is structurally tilting returns toward analog and power-semantics suppliers rather than pure-play GPU manufacturers; higher rack-level power density compresses the fraction of BOM spend that is GPUs and expands analog/PFC/PMIC TAM by a multiple over 3–5 years. That creates durable, less cyclical cashflow optionality for names with proprietary process-node analog IP and pricing power in specialized silicon — these businesses can improve margins through design wins even if system OEM volumes grow modestly.

Second-order winners include advanced substrate/packaging vendors and co-pack optics players — adoption of tighter interconnects shifts margin pools away from discrete DRAM/HBM suppliers toward package and interposer specialists, while also raising the bar for thermal and power management vendors. However, the path is non-linear: architecture pivots (e.g., inference efficiency gains or SRAM experiments) and export/geo-policy frictions could compress projected unit demand within quarters, not years, producing sharp re-rating risk for concentration-heavy names.

For portfolio construction, match time horizon to mechanism: buy-through-product-cycle for analog/power (12–36 months) and use short-dated options as event hedges around major AI announcements (days–weeks). The consensus underweights fragility from single-customer concentration and overestimates the speed at which packaging/optical transitions materialize; that mismatch creates asymmetric entry points into selectively defensive analogs versus swingy infrastructure names.

Key Decisions for Investors

• Long AVGO (6–12 month): allocate 3–5% position via shares or 9–12 month call spread (buy ATM+5% / sell +40%) sized so max premium = 1% portfolio. Rationale: capture margin expansion from increased system-level integration; stop-loss -12% (hard) with target +30–40% if design-win cadence accelerates.
• Long MPWR vs Short SMCI (pair, 6–12 month): buy MPWR (2% portfolio) and short SMCI (1.5% portfolio) to express power-IC win-rate without taking broad server OEM exposure. Expect asymmetric upside if power density cycle continues; hedge size to keep net delta <0.5 to limit GPU-volume beta. Target 2:1 reward-to-risk over 12 months.
• Long TXN (12–24 month, income anchor): buy shares for dividend + margin-improvement exposure, position size 2–4% portfolio. Set trailing stop at -10% and take partial profits at +25%; this reduces portfolio volatility if AI capex becomes lumpier.
• Short-dated NVDA hedge (event protection, 1–3 month): purchase 3-month put spread ~3–8% OTM sized to cost <0.5% portfolio as insurance against a concentrated downside shock from an architecture/expectations reset. If nothing occurs, cost is limited; if a negative surprise hits, hedge payoff is >4x premium.