Nvidia Stock vs. Broadcom Stock: A Wall Street Analyst Says Buy One and Sell the Other

Jay Goldberg recommends selling Nvidia (Seaport target $140, implying ~21% downside from $177) and buying Broadcom (target $430, implying ~37% upside from $314). Goldberg flags risks to Nvidia from $27B of cloud service agreements over six years and $40B of equity investments in customers—potentially inflating demand—and competition from custom TPUs, despite Nvidia reporting adjusted earnings up 82% in Q4 and Street EPS growth of ~53% annually to Jan 2028 while trading at ~37x earnings. Broadcom leads in data-center networking and custom AI XPUs (~60% share for custom accelerators), reported AI semiconductor sales +106% in Q1, total revenue +29% to $19.3B and EPS $2.05, with management guiding to ~46% revenue growth in Q2 and Street EPS growth ~66% annually to Nov 2027. Overall, the piece is company-specific analyst debate rather than market-moving news; both names are presented as attractive but with different risk profiles.

Analysis

Broadcom’s design-for-hyperscaler model creates an asymmetric moat: by owning the spec-to-silicon path for custom accelerators and top-of-rack/leaf switching, it captures both bill-of-materials and systems-level pricing power while increasing switching costs for customers who co-design stacks. That dynamic implies margin compounding that is underappreciated by investors focused only on raw AI-accelerator unit shipments — the real value accrues where silicon, switch fabric, and firmware are sold as an integrated solution to scale-out clusters.

Nvidia’s equity-and-cloud-credit strategy has an important second-order effect on demand visibility and pricing discipline: load-building through partner financing reduces short-term sales cyclicality but also blunts the real demand signal and creates a contingent capital impairment if any funded customers reprice or switch to in-house silicon. Separately, hyperscalers’ shift to purpose-built silicon removes high-margin workloads from the GPU TAM and accelerates commoditization in price-sensitive inference workloads within 6–24 months.

Key catalysts to watch are (1) hyperscaler public deployments of non-GPU models at scale, (2) multi-vendor interop wins that lower software switching costs, and (3) any regulatory scrutiny tying equity investments to procurements. These would compress multiples quickly; conversely, broadened CUDA lock-in or a meaningful software portability gap for XPUs/TPUs would re-center value back to GPU incumbency within the same 12–24 month window.

Key Decisions for Investors

• Pair trade (12-month): Long AVGO via a 9–12 month call spread (buy ATM, sell OTM) size = 2% notional of book; finance by selling a 9–12 month NVDA call (smaller size) or buying a cheaper NVDA put spread. R/R: asymmetric upside if Broadcom wins additional hyperscaler design deals; downside limited to premium paid on the spread and potential short-call assignment. Stop/trim if Broadcom wins >2 large new design engagements announced or NVDA reiterates multi-year demand visibility.
• Event short (0–6 weeks): Initiate a tactical short or put spread on NVDA into next earnings if management emphasizes continued customer financing or equity investments as a growth lever. R/R: short-duration hedge against guidance disappointment and margin miss; keep position small (<=1% notional) due to headline risk and high volatility.
• Long selective hyperscaler-exposure (6–12 months): Long GOOGL via a call or outright equity to capture TPU adoption and software stack monetization; size = 1–2% of book. R/R: convex upside if TPU footprint expands materially; tail risks include slower-than-expected migration and software portability breakthroughs that favor GPUs.
• Portfolio convexity play (12–24 months): Buy a cheap, diversified basket of suppliers to custom-AI stacks (e.g., HBM/packaging/test exposure via proxies) through options or selective longs to capture the infrastructure upside regardless of accelerator winner. R/R: diversified capture of secular capex with softer single-asset binary risk; cut exposure if hyperscalers announce fully in-house end-to-end silicon roadmaps.