Breaking

China Reviews Broadcom AI Networking Gear as Domestic Compute Drive Intensifies

September 23, 2026 at 00:22 EDT

Infra & Chips
Policy & Regulation

China is reviewing its exposure to Broadcom gear as Beijing pushes harder to build AI infrastructure on a more domestic technology stack, according to a September 23 Financial Times report. The move puts a spotlight on a less visible layer of the AI race: the Ethernet switches, network interface cards and custom-silicon relationships that connect accelerators inside large AI clusters.

Sept. 23 · China and Broadcom

Inventory Check Meets AI Build-Out

China is assessing exposure to Broadcom gear as it pushes domestic AI infrastructure. The key question is whether local suppliers can replace not just accelerators, but the networking fabric that makes large clusters work.

$16.7B

Q3 AI semiconductor revenue

221%

Year-over-year AI growth

102.4

Tbps Tomahawk 6 capacity

800G

Thor Ultra AI Ethernet NIC

Broadcom AI revenue step-up

$16.7B

Q3 actual

$21.7B

Q4 expected

Each block represents roughly $2 billion, showing the projected increase rather than just naming it.

Why the gear matters

AI accelerator→Ethernet fabric→Large cluster→Model training and inference

A domestic AI stack has to replace the interconnect layer as well as the compute chips.

What builders value

Open Ethernet fabrics can support multi-vendor clusters and reduce reliance on a single proprietary stack.

What the review tests

Local replacements must mature across chips, optics, software and manufacturing before procurement can shift cleanly.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.