Arista 7050X4 Series

Fourth-generation 400G leaf platform reaching 12.8 Tbps with optional MACsec and 128-way ECMP for cloud-native and AI-ready data center leaf deployments.

Arista Products

Arista 7050X4 Series

Arista 7050X4 Series
32x OSFP 400G + 2x SFP+, 12.8 Tbps, 900 ns latency

DCS-7050PX4-32S
Our Price: Request a Quote

Get a Quote

Arista 7050X4 Series
32x QSFP-DD 400G + 2x SFP+, 12.8 Tbps, 900 ns latency

DCS-7050DX4-32S
Our Price: Request a Quote

Get a Quote

Arista 7050X4 Series
32x QSFP-DD 400G + 2x SFP+ (MACsec: ports 1–8, 25–32), 12.8 Tbps, 900 ns latency

DCS-7050DX4M-32S
Our Price: Request a Quote

Get a Quote

Arista 7050X4 Series
48x SFP28 25G + 4x QSFP-DD 400G, 2.8 Tbps, 900 ns latency

DCS-7050X4-48Y-4DF
Our Price: Request a Quote

Get a Quote

Arista 7050X4 Series
48x SFP-DD 100G + 8x QSFP-DD 400G, 8.0 Tbps, 900 ns latency

DCS-7050SDX4-48D8
Our Price: Request a Quote

Get a Quote

See all 8 models below

The Arista 7050X4 Series is a 1RU 400G fixed leaf switch family delivering 12.8 Tbps of non-blocking switching capacity across 32 QSFP-DD 400G ports and 2 SFP+ management ports. Four models cover different transceiver and breakout configurations: the 7050DX4-32 (QSFP-DD 400G), 7050PX4-32 (OSFP 400G), and variants optimised for AI cluster direct-attach where each 400G port breaks out to 4x 100G sub-interfaces for 128 total 100G endpoints per switch. Sub-microsecond store-and-forward latency at 400G matches the 7050X3 latency characteristic at 100G, maintaining per-packet delivery timing consistency across mixed-speed leaf deployments.

Hardware forwarding tables provide 128K MAC addresses, 128K IPv4 LPM routes, 500K IPv6 LPM routes (≤64 prefix length), 64K IPv4 host routes, and 64-way ECMP across 2K groups, sized for cloud-scale leaf operation without the complexity and cost of a switch-router platform. AI/ML-specific hardware includes ECN marking, PFC priority flow control, and configurable per-queue buffers for RoCE v2 operation, enabling the 7050X4 to serve as an AI cluster leaf switch attaching GPU nodes directly at 400G with lossless Ethernet guarantees.

32 Ports of 400G in 1RU

Thirty-two QSFP-DD or OSFP 400G interfaces in a 1RU chassis provide 12.8 Tbps of leaf switching capacity, double the capacity of 7050X3 100G models in the same rack unit count. Each 400G port supports breakout to 4x 100G sub-interfaces, enabling 128 total 100G connections per switch for environments where server or storage attachment remains at 100G while the leaf-to-spine uplinks operate at native 400G.

400G GPU Node Direct Attachment

With hardware ECN marking, PFC, and per-queue buffer configuration for RoCE v2, the 7050X4 attaches GPU nodes directly at 400G and provides lossless Ethernet for RDMA communication between GPUs on different leaf switches. Direct 400G GPU attachment eliminates the ToR-to-leaf conversion hop that was required with previous-generation 100G GPU interconnects, halving the hop count for cross-rack GPU-to-GPU traffic in two-tier AI fabrics.

Sub-Microsecond at 400G

Sub-microsecond store-and-forward latency at 400G, the same latency characteristic as the 7050X3 at 100G, ensures that increasing leaf port speed from 100G to 400G does not introduce additional forwarding delay for latency-sensitive workloads. This is relevant for AI collective communication, financial trading, and distributed storage where per-hop latency accumulates across multiple fabric tiers.

EVPN-VXLAN VTEP

EVPN-VXLAN VTEP support enables the 7050X4 to participate in BGP EVPN overlay fabrics as a first-hop VXLAN encapsulation point, supporting symmetric and asymmetric IRB routing at the leaf tier. Combined with 64-way ECMP and 128K IPv4 LPM routes, the platform handles both overlay encapsulation/decapsulation and underlay routing within a single hardware forwarding pipeline without requiring a separate gateway device.