Are Broadcom's Custom AI Chips the Key to the Future of AI?

Anthropic announced it will use Broadcom-designed TPUs starting in 2027, adding another major customer to Broadcom’s custom AI chip business. Broadcom said the segment generated $8.4 billion in the latest quarter and expects annual revenue to exceed $100 billion by the end of 2027. The article is bullish on Broadcom’s AI chip opportunity, while noting Nvidia remains dominant in GPUs.

Analysis

The market is starting to price a broader architectural split in AI infrastructure: GPUs remain the default for frontier model training and highly variable workloads, while ASICs are becoming the economic choice once workloads stabilize. That shift matters because custom silicon doesn’t just lower compute cost; it raises switching costs and creates de facto capacity reservations years in advance, which can smooth revenue visibility for the supplier and reduce bargaining power for hyperscalers over time. AVGO is increasingly becoming a toll collector on AI compute rather than just a component vendor.

The second-order effect is pressure on GPU utilization economics. If large model operators can shift even a portion of inference and repetitive training to purpose-built chips, NVDA’s addressable mix could skew further toward the most demanding workloads, which preserves its premium pricing but potentially caps share gains in mature clusters. The real risk for the GPU ecosystem is not immediate displacement; it is that the fastest-growing incremental spend gets routed to lower-cost custom silicon, compressing the multiple the market is willing to pay for the whole AI capex chain.

The timeline is important: this is a 12-24 month re-rating story, not a next-quarter earnings event. The bull case for AVGO strengthens if additional large-scale AI customers follow with design wins, while the bear case is that execution delays, workload drift, or model efficiency improvements slow ASIC adoption and push back the expected revenue ramp. In that scenario, the stock can give back gains quickly because the valuation is already leaning on a very steep forward growth curve.

The consensus seems to underappreciate that this is not a pure AVGO vs NVDA winner-take-all trade. The likely outcome is a bifurcated market where AVGO captures standardizable, high-volume compute and NVDA keeps the frontier and flexibility premium. That makes the opportunity less about shorting NVDA outright and more about relative-value positioning around which parts of AI spend are most defensible versus most commoditizable.

Key Decisions for Investors

• Long AVGO into pullbacks with a 6-12 month horizon; upside remains tied to multi-customer ASIC adoption, but size modestly because the narrative already has momentum and multiple expansion is doing part of the work.
• Use a relative-value pair: long AVGO / short NVDA on a 3-6 month basis if ASIC adoption headlines continue; thesis is mix shift toward lower-cost custom silicon and potential multiple compression for the more mature GPU trade.
• For event-driven exposure, buy AVGO call spreads 6-9 months out rather than outright calls; this captures continued re-rating while limiting downside if adoption timelines slip.
• Avoid chasing NVDA on this specific catalyst; keep it as a core AI hold, but expect underperformance versus AVGO in any period where custom-chip design wins dominate the tape.
• If AVGO gaps higher again on follow-on customer announcements, trim into strength and re-enter on consolidation; the best risk/reward is after the market digests the headline rather than at peak excitement.