Broadcom Logs $73B AI Backlog and Sees 74% Surge in AI Chip Revenue
Broadcom exited fiscal 2025 with a $73 billion AI-related backlog, including $20 billion of non-accelerator networking components set for delivery over the next 18 months. The company’s AI semiconductor revenue jumped 74% year-over-year to $6.5 billion, and management forecasts AI revenue doubling to $8.2 billion this quarter.
1. Networking Infrastructure Emerges as AI Bottleneck
As hyperscale data centers deploy AI clusters exceeding 100,000 compute nodes, bandwidth and latency constraints in existing interconnects have surfaced as the industry’s primary bottleneck. Hyperscalers are transitioning from copper links to 800-gigabit optical modules at scale, while 1.6-terabit silicon photonics interconnects are entering qualification with select customers. Broadcom reported that its AI-related backlog reached $73 billion at fiscal year-end, with nearly $20 billion tied to non-accelerator networking components such as lasers, PCIe switches and high-speed Tomahawk series switches. This networking backlog now accounts for almost 30% of the company’s projected AI revenues over the next 18 months, underscoring the material role that optical modules and switch silicon will play in next-generation AI deployments.
2. Institutional Investors Adjust Broadcom Positions
In the third quarter, two major asset managers trimmed their stakes in Broadcom. Segment Wealth Management reduced its holding by 2.3%, selling 4,291 shares to leave a 3.7% weighting in its portfolio, while HM Payson & Co. offloaded 68,968 shares, lowering its position to 6.2% of total assets. These sales freed up roughly $500 million in combined capital for both firms. Meanwhile, other funds including AssetMark and RiverFront added to their positions, with AssetMark boosting its stake by 2.4% (approximately 39,000 shares) and RiverFront more than doubling its holding with an additional 8,566 shares. Overall, institutional ownership remains above 76%, reflecting continued confidence in Broadcom’s dual exposure to AI accelerators and networking silicon.