40G vs 100G Optical Transceiver

The 40G and 100G generations occupy a unique position in the history of data center networking. They are the two speeds that share the same physical form factor—QSFP—while using different lane rates. A 40G QSFP+ module and a 100G QSFP28 module are mechanically identical: the same cage, the same edge connector, the same latch. The 100G module even fits into a 40G port. But one uses four 10G lanes, and the other uses four 25G lanes, and that single difference defines everything else.

This physical compatibility has made the 40G-to-100G transition smoother than most generation upgrades. It has also created a period of coexistence in which the same switch port can operate at either speed depending on the module installed. Many switches sold over the past decade support both 40G and 100G on the same QSFP28 port, automatically detecting the module and configuring the SerDes accordingly. This dual-speed capability has allowed operators to upgrade from 40G to 100G without replacing the switch chassis—only the optics need to change.

But 40G is now a legacy technology. Its lane rate—10G per lane—belongs to the generation before 25G became the standard building block. The 100G QSFP28 interface, built from four 25G lanes, aligns with the 25G access layer and the 200G/400G aggregation layer that followed. As a result, 40G is being phased out of new deployments, and 100G has become the baseline aggregation speed in modern data centers. Understanding both is still necessary, because the installed base of 40G equipment remains large and the migration path from 40G to 100G is one of the most common upgrade projects in networking.

This guide examines both transceiver families in depth: their lane architecture, wavelength plans, reach, power, cost, breakout capability, and the architectural role each plays in leaf-spine and AI fabrics.

1. Same Footprint, Different Lane Rate

The QSFP form factor was originally defined by the SFF-8436 multi-source agreement for 40G Ethernet. When 100G QSFP28 was standardized under SFF-8665, it kept the same physical dimensions, edge connector, and management interface. This backward compatibility was deliberate: it allowed the industry to reuse cages, connectors, and mechanical designs from the 40G generation, and it gave operators a clear migration path that did not require replacing switch hardware.

The electrical difference is in the lane rate. A 40G QSFP+ module operates four SerDes lanes at approximately 10.3125 Gbps each. A 100G QSFP28 module operates four SerDes lanes at approximately 25.78125 Gbps each. The encoding is the same in both cases—64b/66b—and the modulation is NRZ in both cases. The only change is that each lane runs 2.5 times faster.

The physical compatibility has a practical consequence: a 100G QSFP28 module will plug into a 40G QSFP+ port. Whether it works depends on whether the switch port's SerDes can operate at 25G per lane. Older 40G-only platforms cannot run a 100G module at 100G, though some can run it at 40G if the module supports dual-rate operation. Modern QSFP28 ports support both 40G and 100G, which is why the 40G-to-100G migration is often as simple as swapping modules.

2. Lane Architecture: 4×10G vs 4×25G

The most important distinction between the two generations is what each lane represents. In a 40G module, each lane carries 10G—the same lane rate as a 10G SFP+ interface. In a 100G module, each lane carries 25G—the same lane rate as a 25G SFP28 interface.

This alignment with the access layer is what makes 100G architecturally superior to 40G. A 40G QSFP+ port can break out into four 10G SFP+ links, matching a 10G access layer. A 100G QSFP28 port can break out into four 25G SFP28 links, matching a 25G access layer. When the access layer upgraded from 10G to 25G, the aggregation layer had to upgrade from 40G to 100G to maintain the lane-rate alignment.

The lane-rate alignment is not just an engineering convenience. It determines how the entire fabric scales. When the access layer runs at 25G and the aggregation layer runs at 40G, the aggregation layer is built from a different lane rate and cannot natively aggregate the access links. When the aggregation layer runs at 100G, it aggregates four 25G access links per port, and the fabric scales coherently.

3. Wavelength Plans and Reach

The two generations differ not just in lane rate but in the wavelength plans used for single-mode transmission. These plans determine reach, cost, and compatibility with the fiber plant.

3.1 Multimode Variants

Both 40GBASE-SR4 and 100GBASE-SR4 use parallel optics over multimode fiber. Four 850 nm VCSELs transmit on four fibers, and four photodiodes receive on four fibers, all terminated in a single MPO-12 connector. The difference is the lane rate: 10G per lane for 40G, 25G per lane for 100G.

Because the 40G lane rate is lower, it suffers less modal dispersion and reaches farther. A 40GBASE-SR4 module reaches 100 meters over OM3 and 150 meters over OM4. A 100GBASE-SR4 module reaches 70 meters over OM3 and 100 meters over OM4. This is a rare case where the older, slower generation has a reach advantage—and it is the reason some legacy multimode links that ran comfortably at 40G over 150 meters cannot be upgraded to 100G without fiber remediation.

3.2 Single-Mode Variants

The single-mode variants of the two generations use different wavelength plans, and this difference affects compatibility and cost.

The 40G LR4 and ER4 variants use the CWDM wavelength grid at 1271, 1291, 1311, and 1331 nm—the same grid used by 100GBASE-CWDM4. This means the CWDM multiplexers and demultiplexers in the fiber plant are compatible between 40G LR4 and 100G CWDM4, though the module optics differ.

The 100G LR4 and ER4 variants use the LAN-WDM grid at approximately 1295, 1300, 1305, and 1309 nm—a tighter spacing that requires more precise wavelength control. This plan is not compatible with CWDM multiplexers, which is why 100G LR4 and 100G CWDM4 cannot share the same passive optical infrastructure.

The practical implication is that a fiber plant built for 40G LR4 can be reused for 100G CWDM4 with only module changes, but migrating to 100G LR4 requires either different multiplexers or direct point-to-point fiber without passive multiplexing.

4. Power Consumption

Power consumption follows a pattern that is now familiar from every generation transition: the higher-speed module consumes more power per port but less per gigabit.

The comparison is nuanced. The 40G SR4 module is the most power-efficient per port, consuming only 1.0 to 1.5 watts. But it delivers only 40 Gbps. The 100G SR4 module consumes 2.5 to 3.5 watts—roughly 2.3 times the power—while delivering 2.5 times the bandwidth. The power per gigabit is therefore comparable or slightly better for 100G SR4.

The single-mode comparison is more favorable to 100G. A 40G LR4 module consumes 2.5 to 3.5 watts for 40 Gbps, or roughly 63 to 88 mW per gigabit. A 100G CWDM4 module consumes 3.5 to 4.5 watts for 100 Gbps, or roughly 35 to 45 mW per gigabit. The 100G module is nearly twice as power-efficient per gigabit despite consuming more absolute power.

For a switch with 32 QSFP ports, the difference in total optical power is significant. A 32-port 40G switch dissipating 32 × 1.5 W = 48 W from SR4 optics delivers 1.28 Tbps. A 32-port 100G switch dissipating 32 × 3.0 W = 96 W from SR4 optics delivers 3.2 Tbps. The 100G switch uses twice the optical power to deliver 2.5 times the bandwidth—a net improvement in power efficiency.

5. Cost and Total Cost of Ownership

Cost is where the 40G-to-100G comparison becomes most interesting, because the two generations are at different stages of their product lifecycles.

On paper, the 40G module has a lower cost per gigabit for the SR4 variant. A 40G SR4 module at $80 delivers 40 Gbps at $2.00 per gigabit, while a 100G SR4 module at $250 delivers 100 Gbps at $2.50 per gigabit. This apparent advantage is misleading for several reasons.

First, 40G is a declining market. As vendors phase out 40G production in favor of 100G, availability shrinks and prices may not continue to fall. In some cases, 40G modules have become more expensive than equivalent 100G modules because of reduced production volume.

Second, the cost comparison ignores the switch port. A switch with 32 QSFP28 ports delivering 3.2 Tbps costs less per gigabit of switching capacity than a switch with 32 QSFP+ ports delivering 1.28 Tbps, because the switching ASIC, power supply, and chassis are shared across more bandwidth.

Third, the breakout economics favor 100G. A 100G QSFP28 port broken out to four 25G SFP28 links provides four 25G connections for $38 to $100 each—comparable to or lower than the cost of four individual 25G SFP28 modules. A 40G QSFP+ port broken out to four 10G links provides four 10G connections, but 10G is a legacy access speed that is itself being phased out.

The total cost of ownership calculation therefore favors 100G for new deployments, even though the raw module price is higher. The 40G generation is at the end of its lifecycle, and investing in it means investing in a technology that will not be supported indefinitely.

6. Breakout and Aggregation

Both generations support breakout, and the breakout configuration reveals the architectural role of each.

The 40G QSFP+ port breaks out into four 10G SFP+ links—the access speed of the generation it belongs to. The 100G QSFP28 port breaks out into four 25G SFP28 links—the access speed of the current generation—and also supports 2×50G breakout for the 50G access layer that is now emerging.

This breakout flexibility is why 100G has displaced 40G even in roles where raw bandwidth per port is not the primary concern. A 100G port can serve as a single 100G uplink, as four 25G access ports, or as two 50G links, depending on the module and breakout cable used. A 40G port can only serve as a single 40G link or four 10G access ports. The 100G port is simply more versatile.

7. Architectural Roles in the Data Center

Both 40G and 100G are aggregation-layer interfaces. They connect leaf switches to spine switches, or spine switches to super-spine switches. They are not typically used for server access, where the access-layer speed (10G, 25G, or 50G) is matched to the server NIC.

7.1 40G in the Legacy Architecture

In the 10G access generation, 40G QSFP+ was the standard aggregation interface. A leaf switch with 48 × 10G access ports used 4 to 8 × 40G uplinks to connect to the spine layer. The 40G uplinks matched the 10G access lanes in a 4:1 ratio—each 40G port aggregated four 10G access links.

This architecture is now legacy. The 10G access layer has been largely replaced by 25G, and the 40G aggregation layer has been largely replaced by 100G. New deployments rarely use 40G, except in specific cases where the existing fiber plant or switch hardware constrains the choice.

7.2 100G in the Current Architecture

In the 25G access generation, 100G QSFP28 is the standard aggregation interface. A leaf switch with 48 × 25G access ports uses 4 to 8 × 100G uplinks to connect to the spine layer. The 100G uplinks match the 25G access lanes in a 4:1 ratio—each 100G port aggregates four 25G access links.

This architecture is now the mainstream standard for data center networking. The 100G uplink is the natural aggregation point for the 25G access layer, and it is the interface that most leaf switches and spine switches are optimized for.

The pattern is clear: each generation of access speed pairs with an aggregation speed that is exactly four times the lane rate. 10G access pairs with 40G aggregation (4 × 10G lanes). 25G access pairs with 100G aggregation (4 × 25G lanes). 50G access pairs with 200G aggregation (4 × 50G lanes). The 40G-to-100G transition is the middle step in this progression, and it is the transition that accompanied the 10G-to-25G access upgrade.

8. Migration from 40G to 100G

The migration from 40G to 100G is one of the most common upgrade projects in data center networking, and it is simpler than most generation transitions because the form factor does not change.

8.1 Switch Compatibility

Most switches manufactured in the past decade support both 40G and 100G on the same QSFP28 port. The port detects the module type and configures the SerDes lane rate accordingly. This means that migrating from 40G to 100G often requires only replacing the optics, not the switch.

Some older 40G-only switches cannot support 100G operation because their SerDes cannot run at 25G per lane. These switches must be replaced to migrate to 100G. The cost of replacement is offset by the fact that 100G-capable switches provide four times the per-port bandwidth, so fewer ports are needed for the same aggregate capacity.

8.2 Fiber Plant Considerations

The fiber plant migration depends on the variant. A 40G SR4 deployment over multimode fiber may not be upgradeable to 100G SR4 if the links exceed 100 meters over OM4 or 70 meters over OM3. In that case, the options are to migrate to single-mode fiber with 100G CWDM4 or LR4, or to remain at 40G on those specific links.

A 40G LR4 deployment over single-mode fiber can be upgraded to 100G CWDM4 with only module changes, because both use the same CWDM wavelength grid and the same duplex LC interface. Migrating to 100G LR4 requires different multiplexers, because the LAN-WDM grid is incompatible with the CWDM grid used by 40G LR4.

8.3 Phased Migration

Because the form factor is unchanged, the migration can be phased. Operators can upgrade the highest-traffic links first, leaving lower-traffic links at 40G until demand grows or the modules reach end of life. Multi-rate modules that support both 40G and 100G simplify this approach by allowing a single module part number to serve both speeds during the transition.

9. Comparison Summary

10. Selection Framework

11. Market Context and Outlook

The 40G QSFP+ market has been in decline for several years. As hyperscalers and cloud providers completed their migrations to 25G access and 100G aggregation, demand for 40G modules shifted from new deployments to maintenance and sparing of existing infrastructure. Some vendors have announced end-of-life for their 40G product lines, and the number of suppliers offering 40G modules has shrunk.

The 100G QSFP28 market, by contrast, remains the largest segment of the high-speed optical transceiver market by unit volume. It is supported by a broad ecosystem of switch platforms, module vendors, and cabling infrastructure. The 100G CWDM4 variant, in particular, has become the workhorse for single-mode DCI and campus aggregation because it provides 100G over a duplex fiber pair with 2 km reach.

The next transition, from 100G to 200G, is already underway for the highest-density aggregation layers. The 200G QSFP56 module uses four 50G PAM4 lanes, and the 200G-to-4×50G breakout aligns with the 50G access layer. But this transition is not yet universal; 100G remains the mainstream aggregation speed for the majority of data centers, and it will continue to be deployed in large volumes for years to come.

For the 40G generation, the outlook is clear. It is a legacy technology that served its purpose during the 10G lane era and is now being phased out. New deployments should specify 100G. Existing 40G deployments should plan their migration path, taking advantage of the physical compatibility that makes the 40G-to-100G upgrade one of the least disruptive transitions in Ethernet history.

12.Conclusion

40G QSFP+ and 100G QSFP28 are mechanically identical modules that differ in lane rate. The 40G module uses four 10G lanes; the 100G module uses four 25G lanes. This single difference determines their reach, power, cost, breakout capability, and architectural role.

The 40G generation belongs to the 10G lane era. It aligned with 10G access ports, broke out to 4×10G links, and served as the aggregation interface for data centers built around 10G server connectivity. Its multimode reach is longer than 100G's—150 meters over OM4 versus 100 meters—because the lower lane rate suffers less dispersion.

The 100G generation belongs to the 25G lane era. It aligns with 25G access ports, breaks out to 4×25G links, and serves as the aggregation interface for modern data centers built around 25G server connectivity. It is more power-efficient per gigabit, more versatile in breakout configurations, and supported by a broader and more active ecosystem than 40G.

The migration from 40G to 100G is simpler than most generation transitions because the form factor does not change. Most switches support both speeds on the same QSFP28 port, so the upgrade often requires only new optics. The fiber plant considerations depend on the variant: multimode links longer than 100 meters may need remediation, and 40G LR4 deployments can migrate to 100G CWDM4 but not to 100G LR4 without changes to the multiplexing infrastructure.

For new deployments, 100G is the clear choice. The 40G ecosystem is declining, and investing in it means investing in a technology that will not be supported indefinitely. For existing 40G deployments, the migration path is well-defined and can be executed incrementally as capacity demands grow. The physical compatibility that made 40G and 100G share a form factor is what makes this transition one of the smoothest in the history of data center networking.

13.Q&A