Could Broadcom Become the Next Nvidia?

AI revenue surged ~100% to $8.4B in the quarter and Broadcom expects AI revenue to exceed $10B next period, while forecasting AI chip revenue could top $100B in 2027. The company has multiyear XPU partnerships with six major customers, including Anthropic (initial $10B order plus an additional $11B), and deals with OpenAI, Meta and Alphabet, and is seeing strong demand for Tomahawk 6 networking switches. Shares trade at ~29x forward earnings with a market value around $1.5T, suggesting upside potential and a sector-moving positive development for chip/networking suppliers.

Analysis

Broadcom’s move into bespoke AI accelerators and high-performance switching creates a bifurcated competitive structure versus general-purpose GPU incumbents: one side sells scale and ecosystem (software, model optimizations), the other sells custom, sticky solutions where a handful of hyperscale buyers can drive outsized margins. That creates outsized counterparty concentration risk — a small change in one hyperscaler’s architecture or procurement cadence can swing Broadcom’s growth trajectory more sharply than for a diversified GPU vendor.

The most important second-order supply-side dynamic is advanced-node capacity allocation. Both custom XPU ramps and leading-edge GPUs compete for a finite pool of TSMC/ASML-constrained tooling and qualified substrate partners; prioritization, yield curves, and NRE timing will determine who meets demand first. This implies asymmetric delivery risk over 6–18 months: missing a node or slipping a yield inflection could compress near-term revenue and force punitive pricing or extended lead times.

Regulatory, architectural, and software-ecosystem risks are underappreciated. Export controls, model-architecture shifts (e.g., sparsity, on-chip compression), or a migration to federated/edge topologies that favor many low-power accelerators would materially change throughput/value equations. Conversely, networking stickiness and long-term multiyear design wins can produce annuity-like revenue with higher visibility than typical silicon cycles — a key lever to defend valuation if execution holds.

Key Decisions for Investors

• Pair trade (12-month): Long AVGO equity (core position) and hedge with short-dated NVDA calls (3–6 month) — objective: capture 20–35% asymmetric upside in AVGO driven by XPU/switch annuities while protecting vs a near-term market melt-up in GPUs. Position sizing: 3–4% portfolio long AVGO, 30–50% notional short of 3–6 month NVDA calls; stress test for 40% NVDA move.
• Options trade (6–12 month): Buy AVGO 9–12 month call spread (debit) to limit premium and target 2.5x–4x return if delivery and client ramps validate — max loss = premium, target gain if product milestones hit within 9–12 months. Use strikes ~15–25% OTM depending on current vol to get 4:1 skew vs outright calls.
• Event-driven short (3–9 months): Initiate a tactical short or put spread on AVGO if quarterly guidance misses gross-margin or bespoke-customer cadence expectations — set stop to limit loss at 15% adverse move; payoff skew favorable if customer order cadence delays. Monitor supplier/yield disclosures and hyperscaler spending commentary as triggers.
• Long hyperscaler exposure (12–24 months): Buy selected hyperscaler exposure (e.g., META/GOOGL) via equity or call spreads to play increased AI infra spend without single-supplier concentration — expect 15–30% upside capture if multi-node deployments accelerate; hedge with small AVGO short to neutralize idiosyncratic chip-supplier risk.