The AI Infrastructure Stock That Could Make Investors Millionaires Is Hiding in Plain Sight, and Wall Street Isn’t Paying Attention

Broadcom reported fiscal Q3 revenue of $29.6 billion, up 86% year over year, while non-GAAP EPS nearly doubled to $3.32, driven by a 221% increase in AI-chip revenue to $16.7 billion. Fiscal Q4 revenue guidance of $34.8 billion (+93% year over year) fell modestly short of the $35.1 billion Wall Street consensus, contributing to post-results share weakness. Management expects AI revenue to reach $57.6 billion in fiscal 2026, double to $115 billion in fiscal 2027, and potentially reach $230 billion in fiscal 2028, alongside EPS above $30.

Analysis

The market is discounting a crucial distinction: AVGO's custom-silicon opportunity is not a clean substitute for NVDA GPUs, but it is a higher-visibility, customer-embedded revenue stream once a hyperscaler commits to a program. The economic value accrues disproportionately to AVGO if customers move from pilot deployments to multi-generation ASIC roadmaps, because switching costs encompass silicon design, networking architecture, software tooling, and qualified packaging capacity. That supports a higher-through-cycle earnings multiple than commodity semiconductor suppliers, but only if gross-margin resilience proves that AI mix is not being bought through aggressive customer pricing.

Near-term, the marginal catalyst is not another headline AI-revenue beat; expectations already embed exceptional growth. A 1-3 month rerating requires evidence that non-AI semiconductor demand has bottomed and that backlog converts without supply-chain bottlenecks, while the 6-18 month upside depends on additional named or economically identifiable hyperscaler programs beyond the current customer concentration. The key downside is that hyperscaler capex can remain elevated while AVGO's share of that spend falls: internal accelerators may cannibalize merchant networking or shift mix toward lower-margin custom engagements. Falsify a constructive thesis on any combination of AI growth deceleration, gross-margin compression, reduced multi-year supply commitments, or a material cut to fiscal-2027 earnings power.

Consensus may also be too quick to treat AVGO as simply a cheaper NVDA. Custom ASIC ramps are lumpy, concentrated, and exposed to tape-out delays, yield issues, advanced-packaging constraints, and customer program cancellations; these risks warrant a valuation discount until revenue durability is demonstrated across multiple customers. Conversely, if AVGO shows sustained AI growth with stable margins through two additional reporting cycles, the stock's current relative underperformance could reverse sharply as investors shift from quarterly guide comparisons to multi-year earnings visibility.

Key Decisions for Investors

• Accumulate AVGO only on post-earnings weakness or in tranches over the next 1-3 months; target a 12-18 month rerating driven by validated AI revenue durability and stable gross margin. Size modestly because a concentrated-custom-silicon miss can produce NVDA-like downside without NVDA's broader customer diversification.
• Express relative AI infrastructure exposure via long AVGO / short a basket of lower-differentiation semiconductor beta (SOXX or SMH), rather than an outright AVGO short-volatility trade. The pair works if custom-silicon and networking content gains share; exit if AVGO's AI growth materially decelerates versus the semiconductor index or gross margin deteriorates.
• Maintain NVDA as the cleaner broad-based AI compute exposure and view AVGO as a complementary satellite position, not a replacement. Add to AVGO only if upcoming disclosures show additional customer diversification, multi-year purchase commitments, or evidence that supply availability is translating into shipment growth rather than inventory build.
• Set an earnings watch trigger: reduce or hedge AVGO if management signals customer concentration rising, AI revenue growth normalizing faster than expected, or gross margin declining despite AI mix expansion. Those indicators would imply lower incremental economics and undermine the multiple-expansion case before reported earnings fully reveal it.