When people ask “what optics should I buy?” they usually mean one of two things: “What gives me the best density and lowest cost-per-gig in a spine/leaf fabric?” — or — “What keeps our campus network reliable and cheap to operate?” Those are different questions. The right module for a hot data-center leaf switch is rarely the right pick for a multi-building campus core. Below I explain what matters, why, and which families of modules you should favor in each environment — with practical tradeoffs you can act on today.
Short framing: what the environments demand
A spine/leaf fabric is about massive east-west flows, predictable cabling, and tight power/thermal envelopes. The obvious metric is cost per gigabit and port density. Campus networks are about compatibility, PoE, copper convenience, and a mix of access speeds: lots of endpoints, not a single huge flow. Because those operational goals differ, so do the optics you should buy. Industry experience and vendor designs show 25G hosts and 100G uplinks dominate modern data centers, while campus designs still run a mix of 1G/2.5G/10G with selective 25G where needed.
Spine/Leaf: density first, then everything else
If you’re building or refreshing a spine/leaf fabric, plan for 25G host ports and 100G uplinks as the baseline. The 25G SFP28 form factor gives excellent port density at low watts-per-bit; four 25G lanes map directly to a 100G QSFP28 uplink, which keeps cabling tidy and predictable. Vendors and whitepapers from hyperscale folks back this approach — it’s how you scale bandwidth without exploding rack space or power.
For leaf-to-spine links, QSFP28 (100G) remains the workhorse because of breakout flexibility and broad vendor compatibility. Where fabric density demands it, QSFP-DD or OSFP (400G-capable) are sensible futureproofing steps, but only if your switches and power budgets can handle the modules’ higher draw. Don’t buy 400G simply because it’s trendy — buy it when you need the per-slot capacity. Cisco and other vendors document breakout patterns and energy-aware architectures that show this tradeoff clearly.
A few practical details for data centers:
-
Prefer SFP28 for server ports and QSFP28 for uplinks. That combo gives high density and easy breakouts.
-
Be explicit about signaling: many high-density fabrics still use NRZ (25G lanes) for simplicity; PAM4 shows up in higher-rate optics and demands extra thermal headroom and DSP support. Know what your switches expect.
Campus networks: interoperability and operational simplicity
Campus networks are messy in the best possible way: devices from many vendors, PoE requirements, and plenty of short copper runs. That reality favors tried-and-true pluggable families: SFP for 1G, 10GBASE-T or SFP+ for uplinks, and growing use of 2.5G SFP for Wi-Fi 6 AP backhaul. For building aggregation and inter-building links, single-mode LR (10 km) or short DWDM/CWDM pluggables are common — but you’ll want to avoid highly specialized Coherent parts unless the campus links cross long dark-fiber spans. Practical campus choices prioritize ease of spare-parts management and broad vendor interoperability.
Campus procurement tips:
-
Keep a small set of common part numbers for spares across campus closets.
-
Use copper where PoE is needed. Use multimode fiber (OM4) or single-mode LR optics for building trunks depending on distance and future growth.
-
If you run many inter-building links and expect growth, standardize on DWDM channel plans early — migrating later is expensive.
Modulation and signaling: a practical note
You’ll hear “PAM4 vs NRZ” thrown around a lot. The short practical takeaway: NRZ is simpler and more robust; PAM4 doubles spectral efficiency at the cost of more complex DSP and higher power. In a data center, PAM4 typically appears at higher rates (single-lane 100G or 200G/400G scenarios) and requires careful thermal planning. For campus use cases and many internal rack links, NRZ remains perfectly fine and cheaper. Plan based on actual bandwidth needs, not hype.
Power, cooling, and economics you can’t ignore
High-density optics cost more in power and heat. A rack full of QSFP-DD/OSFP cards can change your cooling profile and power planning. That’s why many teams choose a staggered approach: keep server access at 25G SFP28 and consolidate spine capacity only where necessary. Do the math on watts per Gbps and on overall port density — the cheapest optics per unit aren’t always cheapest per deployed fabric gigabit. Vendor resources and architecture notes stress measuring power/thermal impact before large scale rollouts.
Interoperability and testing: the final step
Regardless of the environment, test optics on the exact switch models you run. Vendor compatibility matrices are a starting point; real interoperability tests catch PME quirks, EEPROM issues, and platform-specific auto-negotiation behaviors. If you must use third-party optics for cost reasons, stage them in a lab and document results — it saves messy escalations later.
Quick buying checklist (one small paragraph, high value)
For spine/leaf: buy SFP28 for servers, QSFP28 for uplinks, consider QSFP-DD/OSFP only with verified power/cooling. For campus: stick to SFP/SFP+, use 2.5G where Wi-Fi backhaul demands it, choose LR or CWDM/DWDM for building trunks based on measured distance and growth plans. Test everything on actual platforms before mass purchase.
WOLON product note
WOLON offers a focused line of optics built for these exact use cases: power-efficient SFP28 and QSFP28 modules tested for spine/leaf fabrics, and a broad set of SFP/SFP+ and 2.5G parts for campus deployments. Each module ships with clear datasheets (TX/RX windows, optical budgets, and power draw) and a compatibility matrix for major switch vendors to reduce surprises at turn-up.
