Spine switches
Spine switches are the fabric layer of a leaf-spine data centre: every leaf connects to every spine, so the spine carries high-radix 100G, 400G and 800G ports and deep tables, and the largest sites add a super-spine tier or a modular chassis. The modular chassis on this page are also the ones vendors position as the core of a three-tier data centre. This page is the spine specifically; the leaf switches that face servers have their own page under top-of-rack.
5 vendors · 16 product families · 90 platforms
Arista
7 families Arista positions for this role
- 7050X3Mainstream
Fixed 1RU/2RU 10/25/40/50/100G data-center TOR/leaf and fixed-config spine — Broadcom Trident 3 X.7-based, sub-microsecond cut-through (800 ns), 32 MB Dynamic Buffer Allocation. Runs Arista EOS; QSFP100 cages also run 40G/4x25G/4x10G via breakout; SFP25 cages run 1/10/25G; 128-way ECMP, MLAG, EVPN/VXLAN. Built for high-density 25G/100G dual-rate leaf or collapsed spine.
- 7060CX5Premium
100G data-center leaf/ToR with 400G uplinks on Broadcom Tomahawk-4 (8.8 Tbps) — 56x100G QSFP access + 8x400G QSFP-DD uplink in 2RU, 825 ns latency, 57 MB buffer. Runs Arista EOS; 100G cages also run 25G, 400G uplinks break out to 200/100G; 128-way ECMP, MLAG, EVPN/VXLAN. RoCEv2, DLB, AI Analyzer. The 100/400G-mix member of the 7060X5 series (below the 400G DX5 and 800G DX5/PX5 tiers).
- 7060DX5Premium
Fixed 400G QSFP-DD data-center leaf/ToR (and fixed-config spine) — 32x400G in 1RU (12.8 Tbps) or 64x400G in 2RU (25.6 Tbps), single-chip, 825 ns latency. Runs Arista EOS; QSFP-DD cages also run 200G/100G; 128-way ECMP, 64-way MLAG, EVPN/VXLAN. Built for hyperscale cloud + AI/ML cluster fabrics.
- 7060X5 (800G)Premium
800G data-center AI/ML leaf/spine on Broadcom Tomahawk-4 (25.6 Tbps) — 32x800G in 1RU, both QSFP-DD (DCS-7060DX5-64/-64E) and OSFP (DCS-7060PX5-64E) connector variants, 825 ns latency, 114 MB buffer. RoCEv2, Dynamic/Cluster Load Balancing, AI Analyzer. Runs Arista EOS; 800G cages break out to 2x400/400/200/100/50G; 128-way ECMP, 64-way MLAG, EVPN/VXLAN. The Tomahawk-4 800G tier below the Tomahawk-5 7060X6 flagship.
- 7060X6Premium
Flagship 800G OSFP data-center AI/ML leaf/spine on Broadcom Tomahawk-5 (51.2 Tbps) — 64x800G in 2RU (51.2T) or 32x800G in 1RU (25.6T), 700 ns latency, Etherlink-certified. RoCEv2 + advanced DCQCN, packet spraying, Dynamic/Cluster Load Balancing, AI Analyzer, and Linear Pluggable Optics (LPO) support. Runs Arista EOS; 800G OSFP cages break out to 400/200/100/50G; 128-way ECMP, MLAG, EVPN/VXLAN. Built for back-end GPU/xPU training fabrics and 800G hyperscale spines.
- 7280R3Premium
Deep-buffer (Broadcom Jericho2, 2–16 GB HBM2) fixed 1RU universal leaf / DCI box / edge router spanning 25G access through 400G — FlexRoute Engine for internet-scale BGP (1.4 M+ IPv4 routes), VOQ scheduling, 512-way ECMP, MLAG, EVPN/VXLAN, MPLS LSR/LER. Runs Arista EOS; QSFP-DD/OSFP 400G cages break out to 200/100/50/25/10G; QSFP100 cages run 100/40G. 3.8 µs latency — a routing/buffering platform, NOT a low-latency AI fabric (use 7060X5/X6 there). DC-only (no PoE), no stacking (MLAG for L2 redundancy).
- 7388X5Premium
Compact modular 4RU 400G data-center switch (single 25.6 Tbps Tomahawk-4 packet processor, 8 interface-module slots) for hyperscale cloud and AI/ML leaf-or-spine fabrics. Runs Arista EOS; QSFP-DD/QSFP56 cages; 128-way ECMP, EVPN/VXLAN.
Cisco
5 families Cisco positions for this role
- Catalyst 9500XPremium
High-performance core — 50-400G, Silicon One Q200, up to 12.8 Tbps fabric
- Hyperfabric 6100Premium
AI fabric — 800G OSFP / 400G QSFP-DD, Silicon One G200/Q200, lossless RoCEv2 for GPU clusters
- Nexus 9300Mid-range to Premium
DC leaf/spine — 1-400G, ACI or NX-OS, VXLAN/EVPN fabric
- N9164E-NS4-O
- N9336C-SE1
- N9364E-SG2-O
- N9364E-SG2-Q
- N9396T12C-SE1
- N9396Y12C-SE1
- N9K-C92160YC-X
- N9K-C92348GC-FX3
- N9K-C93108TC-FX3
- N9K-C93108TC-FX3P
- N9K-C93128TX
- N9K-C9316D-GX
- N9K-C93180LC-EX
- N9K-C93180YC-FX3
- N9K-C93216TC-FX2
- N9K-C93240YC-FX2
- N9K-C9332D-GX2B
- N9K-C9332D-H2R
- N9K-C9332PQ
- N9K-C93360YC-FX2
- N9K-C9336C-FX2
- N9K-C9348D-GX2A
- N9K-C9348GC-FX3
- N9K-C9348GC-FX3PH
- N9K-C9348GC-FXP
- N9K-C93600CD-GX
- N9K-C9364C-GX
- N9K-C9364D-GX2A
- N9K-C9372PX
- N9K-C9372TX
- N9K-C9396PX
- N9K-C9396TX
- Nexus 9500Premium
DC modular chassis — line cards, high-density aggregation and core
- Nexus 9800Premium
Modular data-center spine/core switch built on Cisco Silicon One Q200 (deep HBM buffer) in 10RU (N9K-C9804, 4 line-card slots) and 16RU (N9K-C9808, 8 line-card slots) chassis. 400G QSFP-DD line cards with line-rate MACsec for hyperscale leaf-spine and AI/ML fabrics, with a chassis designed for transition to 800G. Runs NX-OS.
Juniper Networks
2 families Juniper Networks positions for this role
- Juniper QFX5120Mid-range
25 GbE data-center leaf and campus distribution fixed 1U switch line. Trident 3 silicon, MACsec on the -48YM variant, Virtual Chassis stacking via uplink ports. Workhorse 25G ToR for non-AI data centers.
- Juniper QFX 5130High-end
400 GbE QSFP-DD spine-and-leaf / high-density 100 GbE ToR fixed-1U switch line. Broadcom Trident 4 across all 3 chassis. EVPN-VXLAN multihoming via ESI-LAG (Virtual Chassis dropped vs QFX5120). First Juniper line to support 400G-ZR coherent optics on the -48C uplinks.
NVIDIA
1 family NVIDIA positions for this role
- Spectrum SN4000Mid-to-high
Spectrum-3 fixed switches — 400 G data-center leaf/spine for traditional Ethernet and lossless RoCE. Predates the Spectrum-X fabric-bundle positioning (which starts at SN5000 / Spectrum-4).
Nokia
1 family Nokia positions for this role
- 7220 IXR D-SeriesMid-range
Nokia's fixed-1U data-center leaf/spine line — 1G management (D1) through 12.8 Tb/s 400G (D5). SR Linux or SONiC, EVPN-VXLAN fabric (no stacking, no PoE). Nokia publishes no public SKU catalogue for DC switches: SKUit produces a partner-ready Design Brief (RFQ packet), not a quote-ready BOM with SKU line-items.
Other roles
The same catalog, by the job the switch does
- Top-of-rack switchesServer-facing leaf switches for the data centre rack
- Campus access switchesWiring-closet switches for users, phones and access points
- Campus distribution switchesAggregation switches between the wiring closets and the core
- WAN edge and branch switchesSwitching for branch offices and the WAN edge
- Small business switchesCompact switches for small offices and small campuses
Create a validated BOM with spine switches
SKUit turns a use case into a complete, orderable bill of materials — the right platform, optics, licences and power, reconciled against the same validated data shown on this page.
To create a bill of materials for spine switches, you need to log in or create an account.