Bernstein reiterates Nvidia stock rating on AI chip demand outlook By Investing.com

Nvidia projects $1.0T cumulative revenue from Blackwell and Rubin for 2025–2027, with CEO Jensen Huang expressing upside to that figure and noting material contributions from Groq and a potentially 50% larger Vera Rubin generation. Analysts updated targets and ratings (Bernstein/SocGen reiterated $300 Outperform; Rosenblatt raised to $325 Buy; Wolfe $275 Outperform; Barclays $275 Overweight) as Nvidia shows 65% LTM revenue growth and a $4.42T market cap. Management commentary on copper vs optical interconnects, Rubin/Rubin Ultra options and scaling (NV144→NV1152) signals increased demand for interconnect materials and affected related suppliers and China-linked AI names.

Analysis

Nvidia’s roadmap cadence magnifies winners beyond GPUs: advanced packaging and co‑packaged optics vendors, high‑density rack integrators, and specialty memory/power suppliers will see asymmetric demand as customers optimize rack-level throughput and power. Expect incremental BOM shifts — not an immediate rip‑and‑replace — where 12–36 months of design wins can reallocate a nontrivial share of switch and NIC spend from traditional copper PHYs to optics/CPO and associated thermal/power subsystems, boosting revenue visibility for component specialists while compressing legacy copper part cycles.

The main near‑term risk is cyclicality driven by inventory and hyperscaler pacing: a 2–4 quarter inventory digestion at hyperscalers or a model efficiency step (software that halves FLOP demand per task) would materially lower hardware intake and compress ASP leverage. Geopolitical export controls and TSMC capacity tightness remain year‑plus shocks that can either accelerate pricing power (if supply tightens) or force order pushouts (if logistics block distribution), creating asymmetric outcomes across 6–24 month horizons.

Consensus underprices two structural offsets: (1) hyperscaler vertical integration — custom accelerators and network disaggregation can capture targeted workloads without scaling general‑purpose accelerators dollar for dollar, and (2) margin mix risk as lower‑margin rack components (memory, optics) scale faster than high‑margin chips. Trade framing should therefore buy durable share gains while hedging execution and macro risks via time‑staggered options or relative short exposure to vulnerable incumbents in CPU/switch markets.

Key Decisions for Investors

• Buy NVDA 18‑24 month LEAPS (near‑ATM) and fund by selling 3‑6 month near‑ATM calls (calendar debit spread). Timeframe: 12–24 months. R/R: asymmetric upside (target 2x+) if adoption continues; max loss = premium paid for LEAPS minus call premium received (~limited to paid net premium).
• Long NVDA equity sized to conviction and buy a 12‑month 25% OTM protective put (cost = insurance). Timeframe: 6–12 months. R/R: retain >100% upside participation while capping downside to ~25% (minus put cost), ideal ahead of major product ramps or earnings windows.
• Add exposure to co‑packaged optics / advanced packaging suppliers (examples: LITE, GLW, APH) on pullbacks. Timeframe: 6–24 months. R/R: asymmetric 30–70% upside if CPO adoption accelerates and design wins convert; downside 20–35% if adoption stalls or order timing slides.
• Pair trade: long NVDA / short INTC (dollar‑neutral) for 6–12 months to isolate AI accelerator share capture vs legacy CPU exposure. Timeframe: 6–12 months. R/R: reduces macro beta and targets relative outperformance; risk is both outperforming if broader data‑center spend re‑accelerates (loss limited to differential movement).