Broadcom Is Dominating the Custom AI Chip Market and Doesn't Get a Tenth of the Press Nvidia Does

Broadcom’s AI business generated $20 billion in 2025 and management said AI revenue could rise to well above $100 billion by the end of next year, implying roughly 5x growth. The article argues that multiple custom-chip deals with Alphabet, Meta, Anthropic, and OpenAI should materially expand earnings, with consensus estimates calling for 41% annual EPS growth over the next 3-5 years. The stock has already rebounded to an all-time high, but the piece remains constructive on further upside.

Analysis

The market is starting to price AVGO less like a cyclical semiconductor supplier and more like a toll-road on AI capex: once a hyperscaler commits to a custom silicon program, switching costs become embedded in software, networking, validation, and supply-chain tooling. That creates a longer-duration earnings stream than the headline “chip win” suggests, and it also partially insulates AVGO from the GPU pricing cycle that can compress NVDA margins when supply loosens.

The second-order implication is a broader share shift in AI infrastructure spend away from generic accelerators and toward heterogeneous compute. If custom XPUs keep scaling, the real pressure point is not just NVDA share, but adjacent vendors tied to standard rack architectures, merchant networking, and some premium server integrators that rely on homogeneous GPU pods. On the beneficiary side, GOOGL and META likely improve AI capex efficiency per inference dollar, which can expand their tolerance for continued spend even if ad growth moderates.

The market may still be underestimating execution risk on timing, not demand. A multi-customer ramp can create lumpy revenue recognition, packaging constraints, and design-to-volume slippage over the next 2-4 quarters; if any one of the largest programs slips, the stock can de-rate quickly because expectations are now stretched to an aggressive 12-18 month inflection. NVDA is not “hurt” in absolute terms, but the narrative that every AI dollar must flow through GPUs is now broken, and that matters for multiple expansion more than current-year EPS.

Contrarian view: the valuation looks high on trailing metrics but may still be cheap if the AI mix shift proves durable, because the market is applying a single-chip cyclical multiple to what could become a recurring platform royalty stream. The consensus risk is underappreciating how much of AVGO’s upside is already tied to a handful of customers and how quickly sentiment can reverse if those customers prioritize internal platform flexibility or if AI capex pauses for even one budget cycle.

Key Decisions for Investors

• Go long AVGO on any 3-5% pullback over the next 2-6 weeks; target a 6-12 month horizon where a successful XPU ramp can justify further multiple expansion, but size conservatively because expectations are now crowded.
• Maintain a relative-value long AVGO / short NVDA pair for 3-9 months if you want to express AI spend shift without taking outright market beta; AVGO benefits from customization and lock-in while NVDA remains more exposed to cycle and competition risk.
• Buy GOOGL and META on weakness as indirect beneficiaries of custom silicon economics over the next 6-12 months; lower inference cost should support incremental AI deployment and protect free-cash-flow margins.
• Avoid chasing AVGO after sharp upside gaps; use call spreads instead of outright stock if you expect another leg higher, since the main risk is not demand collapse but a timing miss on ramp execution.
• For hedged exposure, pair long AVGO with a small short basket of merchant AI infrastructure names that depend on standardized GPU deployments over the next 1-2 quarters; this isolates the custom-silicon transition theme.