Från 100G till 400G | Tellusfiber

When is an upgrade right for data centres and operator networks?

100G is still a relevant and stable foundation in many networks. But when multiple parallel 100G links are built out between the same nodes, the ports become a limitation or new AI and HPC environments change the traffic pattern, 400G can provide a simpler path to higher capacity. The crucial factor is not just the line speed. The upgrade needs to work with the platform, fibre infrastructure, power budget, cooling, and the planned migration.

Here we go through when it is reasonable to upgrade from 100G to 400G, which technical choices affect the investment and what should be verified before ordering.

Why are data centres and operator networks moving towards 400G?

Traffic is growing in several areas simultaneously. Cloud services, storage replication, DCI, and AI clusters are increasing the demands for bandwidth between servers, leaf and spine switches, as well as between facilities. It is possible to meet the development with more 100G links, but each additional link uses ports, optics, and cabling, increasing the number of components that need to be documented and managed.

A 400G link aggregates four times the nominal capacity of a 100G link in one interface. This does not automatically mean four times better economics or performance. However, 400G can reduce the number of parallel links, free up port capacity, and provide higher bandwidth per rack unit when the platform and traffic model support it.

400G is about more than capacity

A good upgrade decision assesses the cost per transferred bit and the total solution, not just the price of the transceiver. Switch ports, cabling, power consumption, cooling, spare parts, installation, and troubleshooting affect the real cost over time.

In a dense data centre environment, fewer links can simplify cable management and provide more capacity per port. At the same time, 400G optics may have higher power consumption per module and impose clearer requirements on airflow and the platform's thermal limits. Therefore, check both watts per module and watts per transferred capacity.

AI and HPC clusters are changing the traffic pattern

Traditional data centre traffic often goes between users and applications. In AI and HPC clusters, large amounts of data are also moved east-west between accelerators, servers, and storage. This makes consistent throughput, low latency, and a well-thought-out leaf-spine architecture central issues.

400G is used in many modern clusters, but speed alone does not solve issues with oversubscription, buffering, load balancing, or poor cable design. Size the network according to the actual workload and also verify requirements for Ethernet or InfiniBand, breakout, FEC, and platform support.

Five signals that 100G should be evaluated against 400G

• Multiple 100G links are used between the same nodes and the link bundles continue to grow.

• Port density, available switch slots, or cable paths limit expansion.

• The traffic forecast shows that existing capacity will soon not handle normal peaks and planned growth.

• AI, HPC, storage, or DCI projects require more east-west capacity or larger uplinks.

• A planned platform upgrade provides a natural opportunity to introduce 400G without a separate migration.

A single signal is rarely sufficient as a basis for decision-making. However, when multiple signals coincide, it is reasonable to compare a continued 100G expansion with a gradual 400G migration.

100G or 400G - compare the entire link

Checkpoint	Question before decision
Capacity	What load and growth should the link carry during its economic lifespan?
Ports and architecture	Can the platform provide the right port configuration, breakout and oversubscription without unnecessary complexity?
Fibre and connectors	Is the installed fibre compatible with the chosen standard, range, polarity and connector?
Power and cooling	Does the port meet the module's power class and can the chassis cool it in the current configuration?
Operation and inventory	How are spare parts, monitoring, testing methods and skills in the operations organisation affected?
Economics	What does the cost per bit and total cost look like compared to more 100G links?

Which 400G transceiver fits the link?

The right 400G optics is determined first by platform, range, fibre and connector interface. QSFP-DD is common in Ethernet platforms for 400G, while OSFP and QSFP112 are found in other switch and accelerator-close environments. Form factors should not be assumed to be interchangeable just because the data rate is the same.

Variant	Typical usage and checkpoint
400G SR8	Short multimode link. Check OM3/OM4, MPO-16, polarity and any breakout design.
400G DR4	Up to 500 m over parallel singlemode in common IEEE applications. Check MPO interface and breakout to 4 × 100G DR1.
400G FR4	Up to 2 km over singlemode and duplex LC in common IEEE applications. Four optical wavelengths are multiplexed in the fibre.
400G LR4	Up to 10 km over singlemode and duplex LC in common IEEE applications. Verify power budget and receiver levels.
400ZR/ZR+	Coherent optics for DCI and metro. Range and compatibility depend on standard, line system, link budget and platform.

Designations, ranges and interfaces may vary between standards, MSA variants and vendor implementations. Always match both link ends against data sheets and platform compatibility matrix.

Do not forget FEC, breakout and interoperability

At 400G, compatibility is more than the module fitting mechanically in the port. Host interface, firmware, FEC mode, optical standard and breakout configuration must be supported at both ends. This is particularly true for mixed platforms or when a 400G port is to be split into four 100G connections.

Therefore, verify part number, coding, software version, FEC, connector, wavelength, transmitter and receiver levels as well as temperature and power class before commissioning. A short compatibility test in the actual platform reduces the risk of late surprises.

Plan a phased migration from 100G to 400G

A complete overhaul is rarely necessary. Start where the capacity gain and simplification are greatest – for example in the spine layer, between core nodes or on heavily loaded DCI links. Retain 100G where capacity is sufficient and use breakout where the architecture and platform support it.

Document a clear target architecture, test environment and rollback plan. Measure link utilisation, error counters, optical levels and temperature before and after the change. This will allow you to determine if the investment provides the capacity and reliability that was the goal.

Checklist before ordering 400G optics

Confirm platform, port type, operating system and software version.
Determine the actual length of the link, fibre class, connector and polarity.
Choose optical standard based on range and architecture – not just on price.
Calculate power budget and check for any amplifiers or passive components.
Verify FEC, breakout, interoperability and monitoring support.
Check the module's power class, thermal requirements and the chassis's port limitations.
Plan testing, spare parts, documentation and recovery before the change goes live.

When is the upgrade right?

400G is the right step when capacity needs, port density and overall operational economics justify the change – and when platform, fibre and operational model are ready. If 100G still provides sufficient capacity and ease of operation, there is no intrinsic value in upgrading sooner than necessary.

The best decision is therefore based on measurement data and a verified link design. Continue to compare 100G expansion with 400G based on traffic forecast, number of ports, power, cabling, compatibility and total cost.

Do you need help choosing 400G optics?

Tellus Fiber helps data centres, operators and municipal networks choose transceivers, cabling and optical components for the actual link. Verify platform, distance, fibre type, connector, desired port mode and breakout requirements. Then we can help you narrow down the right 400G solution before ordering.

FAQ

What services does your company offer?

In terms of capacity, 400G corresponds to four times 100G, but the replacement depends on traffic flows, redundancy,

port support and breakout. Compare the architecture, not just the sum of the line speeds.

Can existing fibre be used for 400G?

Sometimes. It depends on fibre class, length, connector, polarity and the chosen 400G standard. FR4 and LR4 typically use

duplex singlemode, while DR4 uses parallel singlemode and SR8 multimode with MPO interface.

What is the difference between QSFP-DD and OSFP?

Both are used for high data rates, but they are different form factors with different mechanical and thermal

properties. Start from which form factor and module profile the relevant platform supports.

When is coherent 400G optics needed?

Coherent optics are primarily relevant for DCI, metro and transport links where reach, DWDM and

spectral efficiency require it. Always check platform, line system and link budget.

From 100G to 400G