Broadcom AVGO — Why Connectivity Compounds Differently From Compute
The Durability CurveBroadcom (AVGO) sits at a different layer of the AI stack than NVIDIA. NVIDIA captures compute margin. Broadcom captures connectivity margin — and that difference in position produces a different compounding profile.
This is Law I (Bottleneck Migration): as GPU compute becomes more efficient, the constraint shifts to getting data between GPUs fast enough. Broadcom's Tomahawk and Jericho switch ASICs are the physical infrastructure of AI networking.
The ASIC advantage
Broadcom's custom ASIC business (TPU for Google, Trainium for Amazon) is often viewed as secondary to the networking franchise. This is backwards. Each custom chip deepens the hyperscaler relationship and makes switching costs prohibitive. Three more hyperscaler ASIC deals are in the pipeline.
The risk: VMware acquisition integration. If it stumbles, the thesis breaks. But if it holds — and margin recovery data suggests it is — Broadcom is one of the most durable compounders in AI infrastructure.