Broadcom Is 24% Off Its High and Just Unveiled a Custom AI Chip With OpenAI. Time to Buy the Dip?

Broadcom reported fiscal Q2 AI semiconductor revenue of $10.8B, up 143% YoY, and guided AI chip revenue to grow more than 200% YoY this quarter to $16.0B (within total revenue guidance of ~$29.4B, +84%). Broadcom and OpenAI’s newly unveiled Jalapeño inference chip is expected to move from design to deployment in late 2026 as rack deployments ramp. Despite the strong AI trajectory, the article notes the stock trades ~24% below its 52-week high ($495) amid concerns over customer concentration and custom-chip pricing.

Analysis

AVGO is increasingly less a semiconductor beta and more a high-conviction AI plumbing tollbooth. The key market mechanism is not just faster revenue growth; it is that custom silicon converts AI demand into a more durable, higher-visibility design-win annuity, while also widening the gap between chip budgets and customer-level unit economics. That helps hyperscalers and large model builders by lowering inference cost, which can extend rollout cadence for GOOGL and META, but it also shifts bargaining power away from merchant GPUs over time.

The first-order loser is NVDA at the margin if inference mix keeps moving toward application-specific chips; the second-order loser is anyone trading AI capex as a pure scarcity story. If customers can meaningfully replace some general-purpose compute with lower-cost ASICs, the total AI spend pool may grow slower than unit demand, which can compress multiples across the broader AI infrastructure complex even as end-demand stays strong. That creates a tricky backdrop where AVGO can outperform on earnings while the stock still experiences periodic de-rating whenever investors fear a capex air-pocket or concentration reset.

The catalyst path is mostly 1-3 months: next quarter guidance, new customer disclosures, and any evidence that the six anchor accounts are broadening rather than pausing. Over 6-18 months, the bigger question is whether custom inference becomes the default architecture for large-scale AI workloads; if yes, AVGO’s mix and free cash flow should keep improving, but pricing power will remain the swing factor. The thesis is falsified if AI revenue growth decelerates sharply versus guide or if one of the key customers signals a capex normalization.

Contrarian view: the market may be over-penalizing concentration while underestimating how much lower inference cost expands total tokens consumed. In that scenario, AVGO does not need pricing power on each chip to win; it just needs to keep taking share in a larger, faster-turning AI compute stack. The stock is not a no-brainer, but on a risk-adjusted basis the discount looks more like fear of cyclicality than evidence of broken demand.

Key Decisions for Investors

• Build a starter long in AVGO on any post-earnings or AI-trade weakness, sized modestly; expect 1-3 month upside if guidance stays above current growth rates and customer breadth expands.
• Pair trade: long AVGO / short NVDA for a 1-3 month relative-value expression of the custom-silicon shift; this works best if next commentary emphasizes inference deployment rather than training-only demand. Falsify if NVDA demand re-accelerates faster than expected or AVGO shows customer concentration stress.
• Use GOOGL and META as indirect beneficiaries of lower inference costs: accumulate on AI-related pullbacks if the market starts pricing lower opex from custom chips into 2027 margins. This is a 6-18 month margin-expansion trade, not a quick event trade.
• If AVGO gaps higher on commentary, sell 1-2 quarter out call spreads instead of chasing stock outright; the main risk is not business deterioration but multiple compression on any AI capex scare.
• Set a hard watch item on the next guide: if AI revenue growth drops materially below the current >200% trajectory, reduce exposure immediately; that would signal the market is right to discount the concentration risk.