Switch ASIC Scaling
Overview
Switch ASIC scaling refers to the progression of switching ASIC aggregate bandwidth capacity over time, driven by AI workload demands for ever-higher GPU cluster interconnect bandwidth. The consensus roadmap as of 2025:
| Generation | Capacity | Approximate Timeframe |
|---|---|---|
| Current | 51.2 Tbps | 2025 |
| Next | 102.4 Tbps | 2026–2028 |
| Future | 204.8 Tbps | 2028–2030 |
The Broadcom Tomahawk 6 (sampling June 2025, production early 2026) is the representative 102.4T ASIC, with 2,355B transistors; it enables 64-port 1.6TbE density using 224 Gbps SerDes.
Beyond raw capacity, ASIC scaling introduces significant engineering challenges:
- SerDes evolution: Scaling to 128-port 800G configurations requires migrating from 100G to 200G SerDes per lane, tightening signal integrity constraints on switch PCB design.
- Thermal management: 128-port 800G switches exceed the thermal limits of air cooling; optical transceiver loads at 1.6T speeds can exceed 24W per port, making DLC-cooled 2RU switches a mainstream requirement by 2026.
- Cabling complexity: Each doubling of port bandwidth increases the fiber plant complexity, especially when transitioning from pluggable to CPO or expanded beam optics.
Sign in to read the full article.
Sign In