Broadcom vs. Nvidia: 3 Key Metrics Point to the Stronger AI Chipmaker to Buy After Earnings

Nvidia and Broadcom delivered strong earnings growth and beat consensus, with Nvidia's adjusted revenue and EPS rising 106% and 120% year over year, respectively, versus Broadcom's 86% revenue and 96% EPS growth. Nvidia guided for $108 billion of current-quarter revenue, about $2.5 billion above consensus, and projected 70% fiscal-2028 revenue growth versus analysts' 44% estimate; Broadcom guided for $34.8 billion, slightly below the $35.03 billion consensus, but forecast AI semiconductor revenue of $230 billion in 2028. Both expect gross-margin pressure from higher memory costs and a greater mix of custom chips, but Nvidia is presented as the more attractive investment given its lower forward P/E and $89 billion in quarterly AI-chip revenue versus Broadcom's nearly $21 billion.

Analysis

The key investable signal is not the headline growth rate but the shared margin pressure from memory content. If HBM and advanced-memory pricing remain tight through the next two quarters, Nvidia and Broadcom can preserve revenue growth while incremental gross-margin dollars migrate to Micron (MU), SK Hynix and Samsung; MU is the most liquid U.S. proxy. This also raises the probability that hyperscalers prioritize system-level efficiency and longer platform lifecycles, favoring Nvidia's integrated compute/networking/software stack over standalone custom-silicon deployments.

Broadcom's custom ASIC ramp is strategically important but carries a different earnings quality: customer concentration, lower gross margin and longer deployment/qualification cycles make its out-year targets more vulnerable to a single hyperscaler capex reset. Nvidia's risk is the inverse—its premium economics depend on sustaining rapid platform transitions without customers materially shifting workloads to internally designed ASICs. Over 1-3 months, estimate revisions and HBM supply commentary should matter more than reported beats; over 6-18 months, the decisive datapoint is whether custom silicon captures incremental inference demand rather than merely serving workloads Nvidia would not efficiently address.

Consensus may be underestimating the possibility that both companies win different layers of the same AI buildout, making a directional NVDA-versus-AVGO call less attractive than a quality-versus-input-cost expression. A broad AI multiple de-rating remains the principal near-term risk: if hyperscaler capex guidance softens, neither backlog narratives nor distant revenue targets will prevent multiple compression. Falsify the constructive view if cloud capex plans are cut, Nvidia's gross-margin trough moves below the low-70% range, or Broadcom's AI growth is accompanied by a sustained operating-margin decline.

Key Decisions for Investors

• Maintain a modest long NVDA / short AVGO pair over the next 1-3 months, sized beta-neutral: NVDA has greater near-term estimate visibility, while AVGO is more exposed to long-dated custom-ASIC expectations. Target 10-15% relative upside; exit if AVGO demonstrates AI operating-margin stability and a second major customer ramp.
• Add MU on HBM-supply confirmation or post-earnings pullbacks, with a 6-12 month horizon. The trade monetizes the common memory-cost headwind; reassess if DRAM/HBM pricing rolls over or inventory days rise materially.
• Do not add outright AVGO solely on out-year AI targets until customer concentration, ASIC unit economics and backlog conversion are independently disclosed. Treat hyperscaler capex guidance as the required confirmation catalyst.
• Use NVDA downside hedges around cloud-platform earnings over the next two reporting cycles: put spreads are preferable to reducing core exposure because a capex disappointment could compress the AI complex together even if Nvidia executes operationally.