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

Analyst Jay Goldberg issues a sell on Nvidia (target $140, ~21% downside from $177) and a buy on Broadcom (target $430, below consensus). Nvidia faces scrutiny over $27B of cloud service agreements and $40B of equity investments in customers (Anthropic, CoreWeave, OpenAI) and competition from custom TPUs, even as Q4 adjusted earnings rose 82% and Wall Street models a ~53% annual EPS CAGR to Jan 2028; valuation ~37x. Broadcom, a leader in high-speed networking and custom AI XPUs (≈60% market share for custom accelerators), saw AI semiconductor sales +106% in Q1, total revenue +29% to $19.3B, EPS $2.05, management expects Q2 revenue growth to accelerate to 46%, and Wall Street forecasts ~66% annual EPS growth to Nov 2027; valuation ~43x.

Analysis

The structural move toward bespoke AI stacks is creating a bifurcation: general-purpose GPU volume risk versus rising demand for bespoke networking and ASIC integration. That bifurcation benefits vendors who control interconnects, packaging and systems-level orchestration (e.g., switch silicon, co-packaged optics, advanced substrate partners), while pressuring unit economics for universal accelerators as customers optimize for cost-per-inference and power-per-TFLOP.

A material second-order lever is capital linkage between platform vendors and hyperscalers — balance-sheet exposure or revenue guarantees can magnify booms into sudden busts if one sponsored customer slows; this is a 6–18 month convexity risk that can quickly flip supply/demand and inventory provisioning across the stack. Separately, the software migration friction (toolchains, model conversion, retraining costs) is the slow variable that will determine how fast market share can move across architectures.

Regulatory and concentration risks are asymmetric: forced unwind of strategic investments or an antitrust probe into tied-cloud deals would hit the stock with the most customer-financing exposure materially harder than a pure-play infrastructure supplier. Conversely, a faster-than-expected roll-out of custom XPUs across 2–4 hyperscalers would re-rate networking/ASIC suppliers meaningfully within 12 months as unit economics and recurring software licensing kick in.

The market appears to underweight the margin durability of systems-level suppliers who capture recurring software and switch ASPs while overestimating the permanence of GPU share without accounting for multi-year software migration costs. That creates a window to express a directional view with controlled convexity via pairs and options rather than outright concentrated beta.

Key Decisions for Investors

• Pair trade (6–12 month): Short NVDA 0.6x notional / Long AVGO 1.0x notional. Rationale: asymmetric downside if GPU demand reverts vs steady capture of switch/ASIC economics. Risk: NVDA outperforms on model wins; stop-loss if NVDA > +20% from entry or AVGO < -12%.
• Options collar (9–15 month) on AVGO: Buy AVGO 12–18 month LEAP calls (delta ~0.40) funded by selling nearer-dated NVDA calls (3–6 month, delta ~0.25) sized to be cash-neutral. Reward: convex upside to AVGO re-rating; hedge short-term NVDA positive gamma. Monitor implied vol spread; execute when NVDA IV exceeds AVGO IV by >6 vol points.
• Event-driven long AVGO (3 months): Add on any post-earnings pullback >6% or if guidance implies faster mix shift to custom ASICs. Target 25–35% upside to re-rating within 12 months; hard stop at -10%.
• Tactical hedge for portfolio nervousness (30–90 days): Buy NVDA 30–60 day puts (10–15% OTM) sized to cover concentrated AI exposure; cost acceptable as tail insurance given convex downside scenarios tied to customer funding shocks.