A Generational Investment Opportunity: 3 AI Stocks I'm Buying Now

Nebius' core AI revenue rose 802% YoY in Q4 and the company targets a $7–$9 billion annual run rate by end-2026 (up from $1.25B at end-2025) while expanding data-center sites from 2 in 2024 to 7 in 2025 and 16 expected by end-2026. McKinsey projects $7 trillion cumulative data-center spend by 2030 (Nvidia estimates $3–4T annual CapEx by 2030); the author argues Broadcom and Nvidia are well positioned to capture large shares of AI compute, trade cheaply on forward P/E, and, along with Nebius (stock >20% off its ATH), represent attractive buys.

Analysis

Broadcom and Nvidia are not just overlapping beneficiaries of AI capex — they create distinct demand windows that fragment the silicon supply chain. Broadcom’s client-specific ASIC model increases hyperscaler negotiating power (multi-year appliance footprints, predictable unit economics) and shifts margin capture away from general-purpose GPU aftermarket services; this should accelerate spend into interconnect, custom PCB, and HBM suppliers rather than into repeat GPU purchases. Expect downstream suppliers with fixed-cost fabs or long lead times to see earlier booking visibility and outsized FCF volatility in 12–36 month cycles.

Nebius’ Nvidia linkage offers distributional leverage but introduces concentration and execution risk: preferential access to silicon can convert into a sticky cloud revenue annuity if Nebius secures multi-year commitments, but its capital intensity makes it sensitive to one or two contract misses. Near-term catalysts that will reprice the stock are large hyperscaler contract disclosures, quarterly utilization cadence, and any change in Nvidia supply cadence or shipping priority; negative surprises on any of those can produce >30% re-rating within weeks.

From a portfolio construction standpoint, prefer asymmetric option structures to direct longs here — high conviction on secular adoption but uncertain cadence and convex downside from oversupply, regulatory action, or in-house hyperscaler moves. Contrarian signal: market fear around early-stage cloud builders is likely overdone relative to the probability of at-scale backlog conversion, while the market remains complacent on execution risk at incumbent CPU/accelerator providers that have failed to capture consistent enterprise traction.

Key Decisions for Investors

• AVGO — Buy stock or buy 9–18 month call spread (long ATM / sell 1.2x strike) size 1–3% NAV. Rationale: capture ASIC win-rate re-rating with funded expansion; target +40–60% upside in 12–24 months if a string of hyperscaler deployments convert. Hedge: stop-loss at -18% or sell protective 6–9 month puts to finance premiums.
• NVDA — Buy 12–24 month diagonal call spread (buy longer-dated nearer-ATM call, sell nearer-term higher strike) to capture continued GPU pricing power while limiting IV drag. Risk/reward: expect 2:1–3:1 upside if training demand stays healthy; downside limited to premium paid (10–20% of allocation).
• NBIS — Small-sized equity position (0.5–1.5% NAV) paired with 6–9 month protective puts sized to cap drawdown at ~30%. Timeframe 6–18 months: if utilization and long-term contracts print, equity could 2–4x; if demand stalls, downside is curtailed by puts.
• INTC vs AVGO pair — Short INTC equal notional / long AVGO to express conviction that custom ASIC vendors will out-execute integrated CPU-centric incumbents over 12–24 months. Risk/reward: asymmetric — potential 25–40% relative outperformance vs 15–25% adverse move if Intel regains product momentum; trim on INTC beat + guidance revision.