- Broadcom’s new Trident 5-X12 Ethernet switch is designed for data centres facing growing AI-related bandwidth demands.
- The chip supports 12.8Tbps throughput and integrates software-defined telemetry and energy-saving features.
What happened: Broadcom unveils a new high-performance Ethernet switch to help data centres handle surging AI workloads more efficiently
Broadcom has introduced its latest Ethernet switch chip, the Trident 5-X12, aimed at addressing the rising bandwidth and energy demands created by AI-driven applications in data centres. The chip delivers 12.8 terabits per second (Tbps) of throughput and supports up to 80 100G ports, making it suitable for next-generation leaf/spine network architectures.
Announced on 4 June, the Trident 5-X12 incorporates software-defined telemetry, enabling real-time network insights, and supports network virtualisation, AI workload identification, and dynamic power management. It is built on Broadcom’s latest 5nm process node, offering double the bandwidth of its predecessor while using 25% less power per gigabit.
Broadcom says the chip is suited for enterprises deploying virtualised workloads and AI applications. Ram Velaga, senior vice president and general manager of the Core Switching Group at Broadcom, described the product as “purpose-built” for environments requiring visibility and efficiency at scale. The product is part of Broadcom’s continued investment in Ethernet switching to meet modern data traffic challenges.
Also Read: Broadcom unveils faster custom chip tech to meet GenAI demand
Also Read: Broadcom shares plunge as AI hopes fade
Why this is important
This launch signals Broadcom’s intent to reinforce Ethernet’s role in data centres, even as AI clusters often lean towards custom interconnects like Nvidia’s InfiniBand. As AI workloads become more mainstream, traditional data centre infrastructure is under pressure to evolve. Ethernet remains attractive for its openness and cost-efficiency, but it must adapt to support higher bandwidth, lower latency, and dynamic telemetry.
Trident 5-X12’s software-driven approach and power efficiency offer a counterpoint to more proprietary solutions. While InfiniBand still dominates in high-performance AI training environments, cloud providers are increasingly experimenting with Ethernet-based fabrics to scale inference workloads and reduce costs, as seen with Microsoft Azure’s AI infrastructure updates.
Broadcom’s announcement also coincides with growing scrutiny over energy consumption in AI data centres. By focusing on dynamic power scaling, the Trident 5-X12 addresses rising concerns around sustainability and operating costs. However, the shift to Ethernet for demanding AI tasks remains a work in progress. For now, Broadcom’s latest chip may find strong uptake in hybrid environments where cost, efficiency, and visibility are top priorities.