TBC Home →
← Back to stack

Compute

SubstituteCore: AI training, Consumer / IoTHigh: AI inference, Robotics, EV / auto, Industrial

Connectivity / networking chips

China has strength in telecom systems, but AI datacenter networking and high-speed interconnect remain difficult frontier areas.

The simple version

Connectivity chips move data between devices, servers, sensors, and accelerators. In AI clusters, networking is part of the computer, not an accessory.

What the technology does

  • This category includes Ethernet switches, SerDes, optical interconnect, Wi-Fi, Bluetooth, cellular, and specialty communication chips.
  • For AI clusters, the bottleneck is moving data between thousands of accelerators without wasting time or power.
  • For robots and factories, reliability and latency often matter more than headline bandwidth.

Market scale

  • Broadcom, Marvell, Nvidia, Intel, and Qualcomm are global references across datacenter, optical, wireless, and edge connectivity.
  • Broadcom and Marvell show how networking chips can become strategic AI infrastructure, not just telecom components.
  • The market spans huge datacenter contracts and tiny consumer wireless chips, so company comparisons can be misleading.
Global references
  • BroadcomUnited States · AVGO
  • MarvellUnited States · MRVL
  • NvidiaUnited States · NVDA
  • IntelUnited States · INTC
  • QualcommUnited States · QCOM

China angle

Huawei is the strategic anchor because it combines telecom, networking, cloud, and systems. ZTE Microelectronics, Maxscend, Montage, and Motorcomm represent narrower slices of the stack.

China references

How we got here

China's telecom strength gave it a deeper base here than in many semiconductor categories. The AI cluster era raises the bar because high-speed datacenter interconnect is a different battlefield from ordinary connectivity.

  1. 2000s

    Mobile and broadband growth scale the connectivity chip market.

  2. 2010s

    4G/5G and cloud datacenters increase the strategic importance of networking silicon.

  3. 2022-2024

    AI clusters make Ethernet, optical links, and high-speed interconnect central to compute performance.

  4. Next phase

    Watch domestic AI cluster networking, optical modules, and high-speed SerDes progress.

Why it matters

AI clusters are networks as much as they are chips. Robotics and edge systems also need reliable wireless and wired links.

Main bottleneck

High-speed SerDes, Ethernet switching, optical interconnect, software stack, and global standards ecosystems.

Where this sector shows up

Core
AI training, Consumer / IoT
High
AI inference, Robotics, EV / auto, Industrial

What to watch

  • 800G/1.6T interconnect
  • CPO progress
  • Domestic AI cluster networking
  • Wi-Fi/Bluetooth/UWB adoption