Broadcom BCM87760 Ethernet Switch Chip (Tomahawk 5)

The Broadcom BCM87760, marketed under the Tomahawk 5 family, represents Broadcom's latest generation of high-radix Ethernet switch silicon engineered for hyperscale datacenter and AI/ML fabric deployments. Built on advanced TSMC process technology, the device delivers 51.2 Tbps of full-duplex switching capacity in a single monolithic chip, supporting 512 ports of 100 GbE or scaled configurations up to 64 ports of 800 GbE. The architecture integrates Broadcom's next-generation forwarding engine with deep packet buffers, advanced traffic management, and hardware-accelerated telemetry to meet the demands of lossless, low-latency interconnects required by GPU clusters and distributed AI training workloads.

Tomahawk 5 introduces native support for 800G serdes technology alongside full backward compatibility with 100G, 200G, and 400G interfaces, enabling phased fabric upgrades without forklift replacements. The chip incorporates Broadcom's FlexPort technology, allowing operators to reconfigure port breakout and speed configurations at runtime to adapt to evolving topology requirements. Integrated hardware support for RDMA over Converged Ethernet (RoCEv2), Priority Flow Control (PFC), and Explicit Congestion Notification (ECN) ensures deterministic, low-latency transport essential for east-west AI training traffic and storage fabrics.

Designed for deployment in top-of-rack, spine, and super-spine switching roles within Clos and fat-tree network architectures, the BCM87760 supports comprehensive Layer 2 and Layer 3 forwarding including VXLAN, MPLS, EVPN, and segment routing. The device exposes a standards-compliant management plane through industry-standard APIs and integrates with Broadcom's DNX software development kit, enabling switch vendors and hyperscalers to implement custom control planes. Its power envelope and thermal profile are optimized for high-density chassis and 1U/2U open networking platforms targeting AI datacenter buildouts in enterprise, cloud, and telecommunications environments.

Ideal for

• AI and machine learning cluster fabric switching, providing the lossless 800G interconnect required between GPU and accelerator nodes in large-scale distributed training environments
• Hyperscale datacenter spine and super-spine aggregation, delivering 51.2 Tbps capacity to support massive east-west traffic volumes between compute and storage pods
• High-performance storage networking over RoCEv2, leveraging hardware PFC and ECN support to enable NVMe-over-Fabrics and disaggregated storage architectures with sub-microsecond latency
• Open networking platform deployments where switch builders and hyperscalers require programmable silicon with FlexPort reconfigurability and SDK-level control plane flexibility
• Telco and cloud provider core switching requiring MPLS, segment routing, and EVPN support at line rate across 400G and 800G port densities
• Network performance monitoring and analytics infrastructure, utilizing the chip's integrated hardware telemetry and INT (In-band Network Telemetry) capabilities for real-time visibility into datacenter traffic flows

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.