Catalyst vs Nexus: "Campus" and "Data Centre" Are Labels. These Are the Real Differences.

Written and maintained by Haink's network infrastructure team · Verified against Cisco data sheets, 23 August 2026 · authorized-channel, serial-verified

The received wisdom — Catalyst for the campus, Nexus for the data centre — is correct and useless, because it does not tell you what to do about the server room in the campus building, or the twelve racks that are not quite a data centre. The useful version is that four specific capabilities differ, and if you need any of them the answer is Nexus regardless of what the room is called.

1. Buffers, and why they decide it

This is the difference that has nothing to do with marketing and everything to do with physics. A Nexus 9300-FX carries a 40 MB system buffer. Campus switches are built to a fraction of that, allocated per ASIC.

It matters because of traffic shape. Campus traffic is many clients pulling modestly from a few places. Data-centre traffic includes incast — twenty servers answering one request in the same microsecond, all aimed at one port. The port can only send one packet at a time; everything else queues or dies. Deep buffers absorb that burst. Shallow buffers drop it, TCP backs off, and a storage or distributed-compute workload slows down for reasons no interface counter explains clearly.

So the honest test is not "is this room a data centre." It is: does traffic arrive in synchronised bursts from many sources to one destination? Distributed storage, backup consolidation, MapReduce-style compute and virtualisation clusters all do. Desks and access points do not.

2. vPC versus StackWise

Both solve the same problem — using two upstream switches at once without spanning-tree blocking a path — and they solve it in opposite ways.

The single control plane is exactly what makes StackWise pleasant to operate and exactly what makes it unsuitable under a workload that cannot tolerate a coordinated event. vPC is more work to run and fails in smaller pieces. In a campus closet the first trade is right; under a virtualisation cluster the second usually is.

3. Fabric: VXLAN/EVPN, and ACI

The Nexus 9300 supports standards-based VXLAN EVPN fabrics including hierarchical multi-site, and the same hardware can run in Cisco ACI mode instead of standalone NX-OS. That dual personality is a Nexus property; a Catalyst switch has no ACI mode.

Both platform families can do VXLAN — the Catalyst 9300 supports BGP EVPN VXLAN under Network Advantage — so this is not a binary. What differs is scale, the operational tooling built around it, and whether the fabric can be controller-managed. If the design involves APIC, the hardware question is already answered.

4. Storage and fabric-extension features

FCoE with N-Port Virtualization and Fabric Extender support — up to sixteen Nexus 2000 extenders per 9300-FX — exist on Nexus and not on Catalyst. If converged storage traffic shares the network, or if the topology depends on fabric extenders, the decision is made for you.

The operating system difference is an operational one

Catalyst runs IOS-XE; Nexus runs NX-OS. Feature lists overlap heavily, and the difference an engineering team feels day to day is the surrounding model: configuration idioms, how features are enabled, upgrade mechanics, and — significantly — the fact that a shop standardised on one has muscle memory, scripts and runbooks that do not transfer.

That is a real cost and it belongs in the comparison. A ten-switch Nexus deployment in a team that runs IOS-XE everywhere else means maintaining two operational practices. Sometimes the workload demands it. Sometimes the honest answer is that a Catalyst 9500 will carry that server room and the second practice is not worth it.

Latency

Cisco publishes sub-microsecond latency for the Nexus 93180YC-FX. Campus platforms are not specified to that standard because campus applications are not sensitive at that scale — the difference between one microsecond and five is invisible behind a wireless link, and decisive in a low-latency trading or storage fabric.

Worth being blunt: for most enterprise workloads latency is not the reason to choose Nexus. Buffers are. Latency figures get quoted in comparisons far more often than they change an outcome.

The case that actually comes up

Not a data centre, not a wiring closet — a server room. Twelve to forty racks, virtualisation cluster, some storage, uplinks to the campus core. What goes in it?

Nexus, if the storage is IP-based and shares the network, the virtualisation cluster does live migration at scale, there is any converged storage traffic, a VXLAN fabric is planned, or the cluster's east-west traffic is the dominant flow. All of those are incast patterns and all of them are about buffers.

Catalyst 9500, if the room is mostly north-south, the storage is on its own fabric or is modest, the team runs IOS-XE and nothing else, and the requirement is high-density 25G or 100G aggregation without data-centre semantics. This is a legitimate and common answer, and a supplier who insists on Nexus for every server room is selling rather than advising.

What to establish before choosing

1. Describe the traffic, not the room. East-west or north-south, and is it bursty and synchronised?
2. Is there IP storage on this network? If yes, buffers matter and the answer is likely Nexus.
3. Is a controller-based fabric in the plan? ACI settles it.
4. What does the team already run? A second operating model has a real cost that belongs in the comparison.
5. Do FCoE or fabric extenders appear anywhere? Those are Nexus-only.
6. What are the uplinks? If 400G, see the Nexus 9300 refresh guide — that is a different conversation again.

Frequently asked questions

What is the real difference between Catalyst and Nexus?

Four things: buffer depth, vPC versus StackWise redundancy, data-centre fabric features including ACI mode and FCoE, and the operating system. The Nexus 9300-FX carries a 40 MB system buffer, which is the difference that decides most real cases.

Can I use a Catalyst switch in a data centre?

For north-south traffic without IP storage or heavy east-west bursts, yes — a Catalyst 9500 is a legitimate aggregation switch. Where incast patterns exist, shallow buffers cause packet loss that is hard to diagnose and easy to blame on something else.

Does the Catalyst support VXLAN?

The Catalyst 9300 supports BGP EVPN VXLAN under Network Advantage. What it does not have is ACI mode — that is Nexus-only — and the fabric scale and tooling differ.

What is the difference between vPC and StackWise?

StackWise merges switches into one logical device with a single control plane. vPC keeps two independent control planes that present a shared port-channel. vPC is more work to operate and confines a failure to one peer; StackWise is simpler and shares a fault domain.

Do I need Nexus for a virtualisation cluster?

Usually, if the cluster does live migration at scale or the storage is IP-based on the same network. Both produce synchronised bursts that deep buffers absorb and campus buffers drop.

Is NX-OS harder to run than IOS-XE?

Not inherently, but it is different, and a team standardised on one carries a real cost adopting the other — scripts, runbooks and habits do not transfer. That cost belongs in the comparison alongside the hardware.

Related

• Nexus 9300 refresh — lifecycle position and the 400G question
• Cisco vs Juniper switching — the same decision across vendors
• Leaf-spine networking — the architecture Nexus is built for
• Catalyst 9200 versus 9300 · 9300 versus 9300X
• Enterprise switches · Networking overview · Cisco stock and lead times
• Telecom data hall — 400 kW, 12 racks, three-week turnkey deployment

Sources

• Cisco — Nexus 9300-FX data sheet (40 MB buffer, latency, VXLAN EVPN, vPC, FEX, FCoE NPV, ACI mode)
• Cisco — Catalyst 9300 data sheet (BGP EVPN VXLAN under Network Advantage, StackWise)
• Cisco — Catalyst 9300 architecture white paper