Ethernet Switch Speeds Explained

SONiCopen networkingdata centerAI fabricEthernetautomation

In brief

Engineering guide for campus switching teams covering PoE, access design, MC-LAG/STP risk, telemetry, support, and pilot validation.

Key takeaways

Engineering guide for campus switching teams covering PoE, access design, MC-LAG/STP risk, telemetry, support, and pilot validation.

Why Port Speed Alone Does Not Tell You What Switch You Need

Ethernet switch marketing leads with headline port speeds: 1G, 2.5G, 5G, 10G, 25G, 100G, 200G, 400G, 800G. But a 100G port in a campus aggregation closet solves a completely different problem than a 100G port on a data center leaf connecting GPU nodes. Choosing the wrong tier wastes budget and creates architectural debt that is expensive to unwind.

This guide maps Ethernet speed categories to the workloads Australian enterprise and data center buyers actually run, then shows where open SONiC-based switching can simplify operations across the full stack. Every speed tier discussed here corresponds to a real product category in the xSONiC portfolio, from campus PoE access to 800G AI fabric spine switches.

The Ethernet Speed Tiers: A Practical Map

The IEEE 802.3 working group defines Ethernet speeds from 10 Mbps to 800 Gbps, with 1.6 Tbps work underway. For practical network design, buyers can group these into five operational tiers:

Tier 1 - Campus Access (1G / 2.5G / 5G / 10G): These speeds serve end-user devices, wireless access points, IP cameras, and PoE endpoints. The 2.5GBASE-T and 5GBASE-T standards (IEEE 802.3bz) were specifically created to run over existing Category 5e and Category 6 cabling, making them a cost-effective upgrade path for Australian offices and campuses that cannot justify full cabling replacement. Power-over-Ethernet capability at these speeds is a key differentiator for campus switches.

Tier 2 - Campus Aggregation and Distribution (10G / 25G / 40G): Aggregation layers consolidate traffic from access switches before handing off to the core or data center. In Australian enterprise campuses, 10G SFP+ uplinks remain common, while 25G SFP28 is gaining ground for higher-density aggregation and for connecting to data center fabrics. MC-LAG and STP are common resilience mechanisms at this tier.

Tier 3 - Data Center Top-of-Rack / Leaf (25G / 100G): The 25GBASE-SR and 25GBASE-CR standards dominate server-facing ports in modern data center leaf switches, while 100G QSFP28 uplinks connect leaf switches to spine switches. This is the sweet spot for enterprise data centers running virtualized workloads, container orchestration, and storage fabrics. SONiC has been production-hardened at this tier by some of the world’s largest cloud providers, as noted by the SONiC Foundation.

Tier 4 - Data Center Spine and DCI (100G / 200G / 400G): Spine switches aggregate leaf traffic at 100G or 400G per port. The 200GBASE and 400GBASE standards enable high-radix spine fabrics that reduce hop count and latency. For Australian data center operators managing sovereign workloads, as discussed in OCP Podcast episode 18 featuring Macquarie Data Centres CEO David Hirst, the shift toward AI-class infrastructure is driving demand for 400G spine capacity even in mid-tier colocation facilities.

Tier 5 - AI Fabric and Hyperscale Backbones (400G / 800G): The 800GBASE standard supports the massive east-west traffic patterns of AI/ML training clusters. NVIDIA’s Spectrum-4 and Spectrum-6 switches, for example, deliver 51.2 Tbps aggregate throughput with 64x 800GbE ports in a 2U form factor, and they support SONiC as a network operating system option. This tier is where RoCE v2, RDMA, congestion management, and lossless Ethernet become critical fabric requirements.

Where SONiC Changes the Buying Equation

Historically, each switch speed tier came with a proprietary network operating system tied to a single vendor’s hardware. SONiC (Software for Open Networking in the Cloud) breaks that coupling. As the SONiC Foundation describes it, SONiC is an open-source network operating system based on Linux that runs on switches from multiple vendors and multiple ASICs.

The practical benefit for Australian buyers is threefold:

Multi-vendor hardware flexibility. SONiC runs on switches using ASICs from Broadcom, Marvell, NVIDIA, and others. This means an organization can evaluate switch hardware on price-performance and port density rather than being locked into a single vendor’s NOS licensing model. The OCP Networking project explicitly aims to create fully disaggregated and open networking hardware and software.

Container-based modularity. SONiC’s architecture separates each network function into its own Docker container, enabling independent upgrades of BGP, LLDP, telemetry, or other services without restarting the entire switch. This is operationally significant for teams managing mixed campus and data center environments.

Production-hardened at scale. SONiC has been deployed in some of the world’s largest data centers. The GitHub repository shows active development with thousands of commits and an extensive contributor base. For Australian buyers evaluating SONiC-based switches, this production track record reduces perceived risk.

The key buyer takeaway: when evaluating Ethernet switches across speed tiers, ask whether the NOS supports SONiC or offers a migration path to SONiC. This question matters at every tier, not just in the data center.

Matching Speed Tiers to Australian Workload Profiles

Australian enterprise and data center buyers face a specific set of constraints that influence switch selection:

Sovereign data requirements. The Australian Government’s hosting certification framework and data sovereignty rules mean that more workloads are staying onshore. This drives investment in local data center capacity and, by extension, in campus-to-DC network fabrics that can support growing traffic volumes.

AI infrastructure buildout. As highlighted in the OCP Podcast episode with Macquarie Data Centres, Australian data center design is shifting from cloud-real-estate thinking to chip-out thinking, where rack power density and network bandwidth per GPU node drive architectural decisions. This pulls demand toward 400G and 800G switch tiers.

Campus refresh cycles. Many Australian enterprises are mid-cycle in Wi-Fi 6E and Wi-Fi 7 campus deployments. These wireless standards require multi-gigabit uplinks (2.5G, 5G, 10G) at the access layer and 10G-25G at aggregation, creating a natural pull-through for SONiC-compatible campus switches with PoE.

Geographic distribution. Australia’s geographic spread means branch offices, remote sites, and edge compute locations need right-sized switching. Not every site needs 100G leaf switches, but every site benefits from consistent, manageable NOS tooling.

A Buyer Checklist: Questions to Ask at Each Speed Tier

Before selecting an Ethernet switch at any speed tier, Australian buyers should evaluate the following:

Factor	Campus Access (1-10G)	Aggregation (10-40G)	DC Leaf (25-100G)	DC Spine (100-400G)	AI Fabric (400-800G)
PoE budget per port	Critical	Optional	N/A	N/A	N/A
SONiC or open NOS support	Evaluate	Evaluate	Strong preference	Strong preference	Required for portability
Optical transceiver ecosystem	SFP/SFP+	SFP+/SFP28	SFP28/QSFP28	QSFP28/QSFP-DD	QSFP-DD/OSFP
Latency sensitivity	Low	Low-Medium	Medium	Medium-High	Critical (RoCE/RDMA)
Stacking or virtual chassis	Common	Common	Less common	N/A	N/A
EVPN-VXLAN overlay	Optional	Optional	Recommended	Recommended	Required
Telemetry and visibility	Basic SNMP	Basic SNMP	Streaming telemetry	INT / streaming	INT / IPTPath

The Optical Transceiver Layer: Speed Tiers Need the Right Optics

Ethernet switch port speed is only as good as the optical transceiver plugged into it. Each speed tier has a corresponding transceiver form factor:

1G-10G: SFP and SFP+ modules dominate campus and early data center links.
25G: SFP28 modules for server-facing leaf ports.
100G: QSFP28 modules (4x 25G lanes) for spine-leaf uplinks.
400G: QSFP-DD and OSFP modules (4x 100G or 8x 50G PAM4 lanes).
800G: OSFP and next-generation form factors for AI backbone links.

The transceiver choice affects link distance, power consumption, cabling type (DAC, AOC, or fiber), and total cost of ownership. Australian buyers should evaluate whether their switch vendor supports a broad range of third-party compatible optics or restricts transceiver sourcing. Open networking ecosystems typically allow multi-source transceiver procurement, which can reduce per-link cost and shorten lead times for Australian deployments where specific optics may have limited local stock.

Looking Ahead: 1.6 Tbps and the Multi-Terabit Horizon

The IEEE 802.3dj working group is developing the 1.6 Tbps Ethernet standard, targeting 200G per lane PAM4 signaling. This next generation will further compress AI fabric architectures, enabling higher radix switches with fewer interconnect tiers.

For Australian data center operators planning 3-to-5-year infrastructure roadmaps, the practical implication is straightforward: invest in switch platforms and NOS ecosystems that are actively tracking these standards. SONiC’s open-source development model and its backing by the Linux Foundation give it a structural advantage in adopting new speed tiers as they mature.

The transition from multi-gigabit campus to multi-terabit AI backbone is not a single purchasing event. It is an architectural journey that benefits from consistent operational tooling, open hardware choice, and a NOS that grows with the workload. That is the case for SONiC-based open networking at every speed tier.

Engineering Evidence Floor

For campus and access switching topics, acceptance should be based on endpoint behaviour under real operating constraints. The evidence package should include endpoint classes, PoE budget, NAC/802.1X, LLDP-MED, voice VLANs, multicast, STP or MC-LAG behaviour, uplink capacity, monitoring, rollback, and help-desk workflow. A representative pilot should run at least 48 ports for 30 days, include one planned rollback inside 24 hours, and document support ownership before scale-out.

Evidence area	What to validate	Acceptance gate	Rework trigger
Endpoint mix	APs, phones, cameras, laptops, IoT, and printers	48 ports run mixed load for 24 hours	Only laptop traffic is tested
Power and uplinks	PoE budget, 1G/2.5G/5G/10G access, and 100G uplinks	No power or uplink bottleneck in pilot	Refresh ignores closet constraints
Resilience	STP, MC-LAG, link loss, member loss, and rollback	3 failure cases captured	Campus works only in steady state
Operations	Monitoring, logs, backup, restore, and help-desk runbook	Incident evidence ready within 2 hours	Support depends on informal notes
Rollout	Site selection, training, spares, and change windows	30 days pilot approved before estate rollout	Procurement scales before validation

Engineering FAQ

What should be checked before ordering 400G or 800G optics? Check port form factor, lane speed, reach, fibre type, breakout plan, DOM telemetry, firmware compatibility, thermal budget, and the switch vendor optics support matrix. The same speed can behave differently across QSFP-DD, OSFP, DAC, AOC, and fibre modules.

Why is optics validation part of a SONiC deployment? SONiC exposes the NOS layer, but optics behaviour still depends on the switch platform, transceiver EEPROM data, firmware, thermal design, and operational tooling. Buyers should test the exact module and cable combination before volume rollout.

What should be included in an optics procurement record? Record SKU, reach, connector, fibre type, temperature class, supported breakout modes, switch platform, SONiC image, DOM fields, link test result, and spare strategy. That record becomes the reference for future replacements.

Sources Reviewed

Product fit

Where xSONiC fits

xSONiC can help validate the switch, optics, software image, telemetry, and support assumptions against the actual deployment before a production order is released.