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

Broadcom’s AI semiconductor revenue jumped 143% YoY to $10.8B in Q2, and management guided AI chip revenue to grow more than 200% YoY to $16.0B in the current quarter. Broadcom also highlighted the late-June launch of the Jalapeño inference chip with OpenAI, pointing to increased inference demand from custom accelerators. Despite a 24% discount vs its 52-week high, the article notes strong cash generation ($10.3B free cash flow, 46% of revenue) and a valuation that drops from ~60x trailing earnings to ~19x on forward earnings as AI profits scale.

Analysis

The bigger market implication is not that one chip vendor is winning, but that hyperscalers are increasingly internalizing AI inference economics. That shifts bargaining power away from merchant GPUs and toward custom silicon/IP, which should compress NVIDIA’s long-run pricing power at the margin while improving the cost curves for buyers like GOOGL and META. The second-order winner is the broader AI deployment cycle: lower cost per token should expand inference volumes, so the revenue pool can grow even as unit economics improve.

For AVGO, the key debate is durability versus concentration. The stock is effectively being priced as if a few design wins are enough to support a step-function earnings base, but the real question is whether those wins become repeatable platforms or remain lumpy project revenue; that matters for multiple stability over the next 1-3 quarters. The near-term catalyst is guidance and any commentary on customer ramp timing, while the 6-18 month story depends on whether custom accelerators stay confined to frontier labs or spread into broader enterprise inference.

Contrarian view: the market may be over-fixated on customer concentration and underestimating the stickiness created by nine-month tape-outs and co-development. Once a hyperscaler commits engineering resources and model integration to a custom chip, switching costs rise sharply, which can create quasi-annuity economics even if per-chip pricing is lower. The thesis breaks if AI revenue growth steps down materially from current guidance or if a core customer delays deployment, because then the market will reprice AVGO as a cyclical custom semiconductor name rather than an AI platform compounder.

Key Decisions for Investors

• Buy AVGO on a 3-5% pullback or via a 3-6 month call spread; use it as a moderate-weight AI compounder, not a high-conviction core. Upside comes from multiple expansion if forward growth stays above 60-80%; thesis is falsified if AI revenue growth decelerates sharply next quarter.
• Relative value: long AVGO / short NVDA for 1-3 months as a hedge against AI capex rotation from merchant GPUs to custom ASICs. This expresses the view that custom inference steals mix from GPU vendors while preserving overall AI spend; cover if NVDA re-accelerates or if AVGO customer commentary softens.
• Add GOOGL or META on any AI-infrastructure selloff as indirect beneficiaries of lower inference cost. The trade works if AI capex declines as a percentage of revenue while product engagement/margins hold up; exit if managements signal that custom silicon savings are being reinvested with no margin benefit.
• Set an alert on the next AVGO earnings print: if backlog/customer commentary suggests one anchor account is deferring deployment, reduce exposure immediately. That is the cleanest near-term falsifier because it would hit both growth visibility and the multiple.