Broadcom: Hard To Ignore The Numbers, Buy Before The Market Catches Up

Broadcom's Q3 FY26 revenue surged 85.5% year over year to $29.6 billion, driven by a 221% increase in AI semiconductor revenue and operating leverage that supported margins. Continued XPU demand, expanding deployments, and multi-year hyperscaler partnerships provide strong revenue visibility beyond FY26, underpinning a bullish multi-year earnings outlook.

Analysis

The key investment question is no longer whether AVGO can grow AI revenue, but whether custom-XPU mix can sustain earnings revisions after the market annualizes its current growth rate. Unlike merchant GPU suppliers, AVGO's ASIC programs embed it deeper into hyperscaler roadmaps and create meaningful switching costs through co-design, software integration and packaging qualification. The second-order beneficiary is TSMC (TSM), where advanced-node and CoWoS capacity remain the binding constraint; conversely, each successful internal XPU deployment marginally reduces the addressable accelerator pool for NVDA and potentially pressures its long-duration multiple rather than its near-term revenue.

Near term, AVGO's upside depends on customer deployment timing and supply availability rather than demand discovery. Over the next 1-3 months, any disclosed incremental XPU customer, higher FY27 AI revenue framework, or evidence that networking attach rates are holding alongside compute would support another estimate-reset cycle. Over 6-18 months, the principal risk is hyperscaler concentration: a small number of buyers can defer a program by one quarter, creating large revenue volatility despite a healthy multi-year design pipeline; gross-margin dilution from bespoke programs and rising advanced-packaging costs would be a more important warning signal than headline sales growth.

Consensus may be underestimating the strategic value of AVGO's networking portfolio in an ASIC-led architecture. Custom compute shifts differentiation toward scale-up/scale-out fabric, where Ethernet adoption and higher radix switches can lift silicon content per cluster; this favors AVGO and potentially ANET more directly than a simple "GPU replacement" narrative implies. The contrarian risk is valuation: if the stock has already discounted sustained hyper-growth, merely meeting guidance could compress the multiple even with strong fundamentals.

Key Decisions for Investors

• Maintain/add AVGO on 5-10% post-results or semiconductor-wide pullbacks, with a 6-12 month horizon. Underwrite upside to further FY27 EPS revisions from incremental XPU ramps; reduce if management indicates a material custom-silicon gross-margin step-down or any major program deferral.
• Express the architecture shift as long AVGO / short NVDA in equal dollar beta-adjusted size over 6-12 months, sized modestly because NVDA retains near-term supply and software advantages. The catalyst is hyperscaler evidence of expanding internal-XPU deployments; stop out if NVDA sustains accelerating data-center growth while AVGO's AI backlog or networking growth decelerates.
• Add TSM as a supply-chain complement rather than chase a pure AVGO momentum move. Advanced-node and packaging demand should convert design wins into foundry revenue over 12-18 months; key falsifier is confirmed CoWoS capacity expansion outpacing demand or customer migration to less packaging-intensive architectures.
• Monitor ANET for a read-through trade around hyperscaler capex and Ethernet fabric attach. Initiate only if order commentary confirms AI-networking demand is broadening beyond a limited customer set; avoid if AI compute growth is accompanied by weak switching revenue, which would challenge the cluster-networking thesis.