3 Undervalued AI Stocks to Buy Right Now

Microsoft reported Azure revenue up 39% and overall revenue up 17% YoY, yet the stock is down >30% from its high and near its lowest P/E in a decade, flagged as a buying opportunity. Nvidia trades at 20.6x forward earnings (on par with the S&P 500) while management raised the addressable chip-sales projection to $1 trillion by end-2027, implying significant unpriced growth. Micron trades at ~6.1x forward earnings amid a projected HBM market expanding from $35B in 2025 to $100B by 2028 and current supply bottlenecks; the article recommends accumulation across all three names.

Analysis

The market is treating AI leadership as a short-duration event rather than a structural reallocation of compute and memory budgets; that creates a two- to four-quarter window where demand shock winners (Nvidia, cloud hosts) can out-earn peers and reprice materially while cyclical suppliers (Micron) see headline volatility. Hyperscaler procurement patterns matter more than model count — large, lumpy build-outs take 9–18 months from contract to rack deployment, so expect sequential revenue shocks for suppliers driven by quarterly procurement windows and wafer-start lags rather than smooth linear demand.

Second-order winners include cloud integrators and ISVs that monetize hosting (driving higher ARPU per enterprise) and EDA/equipment suppliers upstream of HBM and HBM-2/3 stacks; second-order losers are commodity DRAM OEMs and legacy CPU vendors that lose share in AI inference/hosting. Geopolitical frictions and export-controls are a persistent regime risk that can bifurcate supply (premium to non-restricted vendors) and accelerate substitutability for customers that can re-architect away from restricted architectures over 12–24 months.

Micron’s cheap multiple embeds a deep cyclical fear but also underestimates structural HBM scarcity — pricing power can persist through 2027 if capex discipline and long lead times keep utilization elevated. Conversely, efficiency innovations (e.g., memory-saving quantization or new compiler/runtime stacks) could reduce per-model memory demand by a non-trivial 10–30% over 12–24 months, meaning Micron’s upside is high-conviction but asymmetric and time-dependent.

For Microsoft and Nvidia the clearest catalyst cadence: quarterly guidance from hyperscalers and chip revenue cadence (next 2–6 quarters) will re-rate multiples faster than macro; watch cloud bookings and HBM ASPs as leading indicators. Positioning should focus on timing these procurement cycles and buying optionality into multi-quarter shipment ramps while hedging for model-driven efficiency and policy-driven export disruptions.

Key Decisions for Investors

• Buy NVDA via 12–24 month LEAPs to express multi-year AI compute dominance: initiate a buy of Jan-2028 LEAP calls (target 0.45–0.65 delta) sized 3–6% notional; finance by selling 6–9 month calls (reduce carry). Risk/reward: limited premium risk (~100% downside to premium) vs potential +50–150% if Rubin/Blackwell revenue inflects over 12–24 months.
• Overweight MSFT with covered-call overlay for 6–12 months: accumulate on 5–10% pullbacks, sell 6-month ATM calls to generate yield and cut basis; target total return 10–25% if cloud multiple re-rates, with upside capped by sold calls and downside protected by premium collected (~3–6% buffer).
• Smaller, hedged Micron exposure: initiate a starter long (2–4% portfolio) and buy 3–6 month 8–12% OTM puts to cap downside, or purchase a put-protected stock position (buy stock + buy 6-month 10% OTM put). Rationale: capture HBM tightness upside while limiting cyclical drawdown risk; expect binary outcomes over 6–18 months.
• Pair trade: long NVDA / short INTC equal notional over 6–18 months to capture share-shift to accelerators — trim if both rally >25% (macroeconomic correlation risk). Risk/reward: levered capture of architecture shift; risk of macro lift that lifts broad semis (use stop at 12–15% adverse move).