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 revenue and CEO Hock Tan said AI revenue could rise to well above $100 billion by the end of next year, implying roughly 5x growth in 24 months. The article argues Broadcom’s custom XPU deals with major AI customers like Alphabet, Meta, Anthropic, and OpenAI could materially re-rate the stock despite its 82 P/E, with consensus earnings expected to grow 41% annually over the next 3 to 5 years. The piece is highly supportive of the stock, though it is more an investment thesis than a new company event.

Analysis

The market is starting to price a structural mix shift in AI silicon, not just another growth cycle for one supplier. Broadcom’s advantage is that custom chips become embedded in customer roadmaps and software stacks, which makes the revenue stream stickier than a typical component win and raises switching costs for hyperscalers over a multi-year horizon. The second-order effect is pressure on the “one-size-fits-all” GPU trade: even if GPU demand stays strong, custom silicon can cap Nvidia’s pricing power in the largest and most cost-sensitive workloads.

The bigger read-through is that AI capex is diversifying rather than slowing. If the largest buyers are simultaneously buying GPUs, networking, and custom accelerators, the risk shifts from “AI demand exists or not” to “who captures the wallet share.” That should benefit the broader AI infrastructure stack, but it may also force Nvidia to defend share with pricing, bundling, or faster product cadence, which could compress near-term gross margin expectations if the market starts to see custom ASICs as a real substitute at scale.

The main risk is timing: these programs are lumpy, engineering-heavy, and can slip by quarters, so near-term numbers may look better or worse than the underlying trajectory. Consensus is probably underestimating how much of AVGO’s upside is already mechanically tied to a handful of very large design wins, which creates a clean bull case but also concentration risk if one customer defers tape-out or delays volume ramp. Conversely, the move may still be underdone because the market is valuing AVGO like a high-quality compounder rather than a platform beneficiary with multiple years of design-win annuities ahead.

Key Decisions for Investors

• Stay long AVGO into pullbacks; prefer staggered entry on 3-5% dips over chasing strength. Risk/reward is attractive if the market continues to re-rate custom AI exposure as a multi-year annuity rather than a one-time cycle.
• Fade relative underperformance in NVDA vs. AVGO via a long AVGO / short NVDA pair over 3-6 months. The thesis is not that GPUs lose share immediately, but that custom silicon progressively dilutes marginal upside in the highest-volume inference workloads.
• Add GOOGL and META on weakness as indirect beneficiaries of custom AI capex; they gain optionality from lower training/inference cost curves and reduced single-vendor dependence. Best expressed as a basket long versus broader tech if you want to isolate AI infrastructure intensity.
• Use AVGO call spreads for the next 6-9 months if implied volatility remains elevated. This captures further multiple expansion while defining downside if design-win monetization slips a quarter or two.