Broadcom Inc. (AVGO)

Broadcom occupies a unique position in the AI stack: rather than selling off-the-shelf GPUs, it designs custom AI accelerators (ASICs) built to the exact specifications of hyperscalers. Google's Tensor Processing Units (TPUs) and Meta's MTIA training chips are both Broadcom designs. This creates deep, long-term customer lock-in at massive scale.

Broadcom also dominates AI networking through its Ethernet chips and switches, which connect the tens of thousands of GPUs inside AI data centers. As clusters scale to hundreds of thousands of chips, networking becomes a critical bottleneck — and Broadcom sits at the center of it.