As hyperscale data centers expand AI clusters beyond 100,000 compute nodes, the industry is shifting focus from raw compute to the critical networking layer. Tens of thousands of optical modules are now required to link GPUs, switches and server racks, and copper cabling can no longer meet the demands for bandwidth, latency and reliability. Adoption of 800-gigabit optical transceivers is accelerating, while 1.6-terabit silicon photonics interconnects have entered advanced qualification. Investors have largely overlooked this networking upgrade cycle in favor of GPU makers, creating an underappreciated growth runway for infrastructure suppliers.

Broadcom exited its fiscal 2025 year with an AI switch backlog exceeding $10 billion, driven by its 102 Tb/s switch bookings, and a total AI-related backlog of $73 billion to be fulfilled over the next 18 months. Nearly $20 billion of that mix is non-accelerator content such as lasers, digital signal processors and PCIe switches, underscoring networking components as an emerging material revenue source. Meanwhile, AI semiconductor revenue grew 74% year-over-year to $6.5 billion in the most recent quarter, and management forecasts that AI revenue will double year-over-year to $8.2 billion in the current quarter.

Institutional investors continue to hold roughly three-quarters of the company’s shares, though some managers have trimmed positions: Segment Wealth Management reduced its stake by 2.3%, and HM Payson sold 5.0% of its holdings, reflecting portfolio rebalancing rather than a thematic shift. Insider selling included approximately 650,000 shares transacted over the past 90 days, representing 2.0% of total insider ownership. Meanwhile, the board increased the quarterly dividend from $0.59 to $0.65 per share, marking a 10% rise and signaling confidence in long-term free cash flow generation.