Broadcom’s Q3 guidance misses expectations by $1.2 billion, plunging over 13% after hours—has the AI narrative “cooled”? - 深潮TechFlow

Broadcom's Q3 AI chip revenue guidance fell $1.2 billion short of expectations, triggering a more than 13% drop in after-hours trading. Despite strong Q2 performance, including a 143% YoY increase in ... Read original →

Broadcom’s Q3 guidance miss signals not an AI slowdown but a structural shift in the tech stack. The drop in AI networking’s revenue share—from 40% to 30%—reflects hyperscalers migrating from generic Ethernet to proprietary interconnects like NVIDIA’s Spectrum-X, directly pressuring optical module vendors (e.g., Innolight, Eoptolink) tied to Broadcom’s switch ecosystem. Tightening U.S. export controls on sub-3nm EUV tools are inflating foundry costs for TSMC and Samsung, raising supply risk for both Broadcom and Marvell. NVIDIA may exploit this by bundling its own switches with GPUs, while SK Hynix could push HBM4-integrated silicon photonics to dominate next-gen interconnect standards. Over the next 12–24 months, the AI chip race will pivot from raw compute to system-level efficiency—marginalizing non-integrated players.