Key Highlights

  • Broadcom's AI semiconductor Revenue of $10.8 billion in Q2 FY2026 encompasses both custom AI accelerators (XPUs) and AI networking silicon.
  • Total semiconductor solutions revenue of $15.009 billion in Q2 FY2026 grew 79% year-over-year, with AI the dominant growth driver.
  • Q3 FY2026 AI semiconductor revenue guidance of $16.0 billion (200%+ year-over-year growth) encompasses both accelerator and networking revenue streams.
  • AI cluster networking creates recurring upgrade cycles as hyperscalers expand AI Training and inference capacity.
  • Analyst models typically aggregate AI revenue without decomposing accelerator versus networking, potentially undermodelling the diversity and durability of the growth stream.

When the Investment community discusses Broadcom Inc. (Nasdaq: AVGO) and artificial intelligence, the conversation is almost entirely about custom AI accelerators — the XPUs designed to train and run large language models for hyperscaler clients. $10.8 billion in a single quarter commands attention. But within that $10.8 billion lies a second, underappreciated category: AI networking.

The silicon that connects tens of thousands of AI accelerators into coherent training clusters — the Ethernet switching chips, network interface controllers, and optical interconnect components — is as essential to AI infrastructure as the accelerators themselves. And Broadcom is among the world's leading suppliers of that networking silicon.

Why AI Networking Is Different

Traditional data centre networking connects general-purpose servers to storage and to the internet. AI cluster networking is fundamentally different. Training a large-scale model on tens of thousands of accelerators requires all-to-all communication at microsecond latency, with bandwidth requirements that scale with model size and chip count. This means AI networking silicon must be purpose-designed for the specific topology of AI clusters — precisely the kind of problem Broadcom's semiconductor design expertise addresses.

The Recurring Upgrade Logic

AI networking creates a structural Recurring Revenue dynamic. As hyperscalers expand AI capacity by adding more accelerators, they must proportionally expand networking infrastructure. Given that Broadcom guided AI semiconductor revenue to grow from $10.8 billion in Q2 to $16.0 billion in Q3 — a $5.2 billion sequential increase — the implied networking component of that increase is substantial.

Moreover, as AI model architectures evolve, networking requirements evolve with them. Mixture-of-experts models, distributed inference, and memory disaggregation all create Demand for higher-bandwidth, lower-latency networking. Each architectural evolution is a potential upgrade cycle, creating a product roadmap with recurring revenue opportunities extending years into the future.

The Analyst Coverage Gap

Most Sell-Side models decompose Broadcom's semiconductor segment into AI revenue (growing fast) and non-AI revenue (growing slowly). Within AI revenue, the further decomposition into accelerators versus networking is rarely modelled explicitly. Investors may be underestimating the diversity within the AI semiconductor category and therefore underestimating the durability of the growth trajectory. If accelerator revenue were to face competitive pressure, the networking Business would provide a partial buffer. If networking grows ahead of consensus, the total AI semiconductor number could exceed current estimates.

Disclaimer: This article is for informational purposes only and does not constitute financial advice or investment recommendation. All data sourced from Broadcom Inc. Q2 FY2026 Earnings release dated June 3, 2026. Past performance is not indicative of future results. Investors should conduct their own Due Diligence.