Switch Chip: AI Networking Innovation’s “Third-Core Hardware”—Can Domestic Substitutes Break the Deadlock?

Switch switching chips are the “neural networks” of AI computing clusters, providing high-speed data transmission and routing functions between GPUs, servers, and I/O devices. In AI large-model training and inference computing clusters, the performance of switching chips directly determines the cluster’s expansion efficiency, communication latency, and overall computing utilization.

The global PCIe switch market is about $2.8 billion and is expected to grow to $6.9 billion by 2034. However, Broadcom holds a 70%+ share in the global commercial switching chip market—against the backdrop of AI computing power becoming a core battleground in great-power competition, the domestic substitution of switching chips is not only a business proposition, but also a strategic security imperative.