I Recommended 2 AI Stocks a Year Ago. Only 1 Beat the Market in 2025, but Both Are Still Worth Buying Hand Over Fist in 2026.

Nvidia remains the dominant supplier of GPUs for AI training and inference and is launching the Rubin platform in 2026, which the author says cuts required processors to one-quarter for training and one-tenth for inference versus Blackwell; analysts are projecting 52% revenue growth for fiscal 2027 and the stock trades at ~40x forward earnings after a 39% 2025 gain. Meta's core social-business accelerated with Q3 revenue up 26% year-over-year, but investor skepticism about sizable 2026 data‑center capex (and whether AI spending will generate adequate returns) has left the stock roughly 15% below its August highs and trading at ~22x forward earnings; the author views both names as buys given Nvidia's secular advantage and Meta's rebound potential.

Analysis

Market structure: Rubin-era NVDA (NVDA) further consolidates winners — Nvidia itself, cloud providers (MSFT, AMZN, GOOGL) and large enterprise AI adopters — while second-tier chip suppliers and bespoke CPU incumbents face margin pressure. Rubin’s 4x training / 10x inference efficiency (vs Blackwell) increases switching costs and pricing power, implying tight GPU demand and limited near-term supply elasticity (12–18 months), with upward pressure on data‑center power/copper demand and higher implied vols in options markets.

Risk assessment: Key tail risks are export controls on advanced GPUs, TSMC capacity shocks, or a sharp enterprise demand pullback; any of these could cut NVDA upside by >30% in 3–6 months. Timeframes: immediate (days–weeks) for earnings/guidance and Rubin shipping headlines, short (3–12 months) for Rubin adoption and Meta (META) capex vs ROI clarity, long (2+ years) for TAM realization and competitive responses from AMD/Intel/custom silicon.

Trade implications: Favor sized long exposure to NVDA (equity or Jan‑2027 LEAPS) and opportunistic long META on valuation dislocation (22x forward vs near‑term capex fear). Use pair trades (long NVDA / short AMD) to express differentiated moat, and use defined‑risk option structures (6–12 month call spreads on NVDA; 9–12 month call spreads or long-dated calls on META) to capture asymmetric upside while capping premium spend.

Contrarian angles: Consensus underestimates how Rubin can entrench Nvidia pricing power and accelerate cloud consolidation; conversely, market may be overpricing Meta’s capex risk — a continuation of mid‑20% ad growth would likely push META from 22x to 26–30x forward within 6–12 months. Watch for hyperscalers accelerating custom silicon programs as the main structural downside that could erode NVDA’s premium over 24–36 months.

Key Decisions for Investors

• Establish a 3% portfolio long position in NVDA via shares or equivalent notional; hedge with 1% notional of 3‑month puts (to limit immediate drawdown). Add to the position on pullbacks to <=35x forward earnings or a price decline of 12% from entry; re-evaluate/take profits if position rises +50% or forward PE expands above 60x.
• Initiate a 2.5% portfolio long in META using a 9–12 month call spread (buy ATM call, sell 35–40% OTM call) to cap cost; add another 1–1.5% if META trades down 10% or forward PE compresses to <=18x. Trim to half if quarterly guidance shows ad revenue deceleration >5% QoQ or capex guidance surprises above market consensus by >15%.
• Implement a relative‑value pair: long NVDA / short AMD (AMD) equal notional at 1.5–2% net exposure to capture moat divergence; close or flip if the NVDA/AMD spread tightens by 30% or AMD posts >20% upside revenue surprise in a quarter.
• Rotate +300bp into semiconductors (SOXX) and cloud software (MSFT, AMZN, GOOGL) funded by -200bp from small‑cap ad/social names (e.g., SNAP). Monitor GPU export control or TSMC capacity announcements over the next 30–60 days and reduce NVDA exposure by 50% if restrictive measures are announced.