Beijing escalates domestic chip drive with Broadcom audits

The Broadcom audit is less a trade action than a structural inflection point where technology sovereignty transitions from policy rhetoric to administrative enforcement. At the silicon level, Tomahawk-class switches and PCIe retimers form the circulatory system of AI cluster interconnects. Domestic substitution will accelerate at mid-range tiers, yet sub-7nm yield constraints keep high-end switch silicon tethered to Taiwan, China foundry capacity — a design-autonomy versus manufacturing-dependency tension unlikely to resolve before 2027. On compliance, the audit mechanism is more penetrating than an outright ban: it forces Broadcom to maintain segregated compliance infrastructure in mainland operations, inflating overhead by 15-25% while creating regulatory hooks for future data-localization mandates. In the competitive chessboard, Marvell and Astera Labs absorb spillover demand, but NVIDIA's NVLink closed-loop architecture already internalizes most network silicon consumption. The structural winners are domestic GPU vendors executing a de-Broadcomization playbook. The 12-24 month thesis: global AI compute infrastructure will fracture along interconnect protocol stacks into two mutually incompatible ecosystems — a deeper rupture than any single-chip export control.