What Is 400G Technology? A Practical Guide
400G is short for 400 gigabit Ethernet: a networking technology capable of moving 400 gigabits of data every second over a single link or wavelength.
That's four times the throughput of 100G, the speed that most data centers and service provider networks have relied on for the past several years. The terms get used loosely, so it helps to separate them:
400G describes the overall technology or the capacity of a link, most often a single optical wavelength carrying 400 gigabits per second.
400GbE (400 Gigabit Ethernet) refers specifically to the Ethernet standard that defines how switches, routers, and network interface cards communicate at this speed.
400Gb/s is simply the raw data rate: 400 billion bits per second, regardless of which standard or transport method delivers it.
400G exists because traffic growth outpaced what 100G links could handle without adding an unmanageable number of ports and cables. AI training clusters, hyperscale cloud platforms, and video-heavy consumer traffic all generate volumes of traffic that make higher per-link capacity a practical necessity, not a luxury upgrade.
How does 400G actually work?
Getting from 100G to 400G required new ways of encoding data onto light. Most 400G links for shorter distances, inside a data center or between adjacent racks, rely on PAM4 modulation (Pulse Amplitude Modulation, 4-level).
Older networking technologies typically used NRZ (Non-Return-to-Zero) signaling, which distinguishes only two amplitude levels per signal pulse: effectively one bit at a time. PAM4 uses four distinct amplitude levels, so each pulse carries two bits instead of one, doubling the data rate without doubling the signal speed.
For longer distances, metro links and data center interconnect (DCI) spanning tens or hundreds of kilometers, coherent optics take over. Coherent transceivers modulate phase, amplitude, and polarization together (using formats like QPSK or QAM) and use digital signal processing (DSP) to keep the signal clean over long fiber runs.
This is more complex than PAM4, and the DSP work adds some power draw and processing latency, worth keeping in mind if latency is a priority for your application.
These standards define how all of this fits together:
IEEE 802.3bs is the Ethernet standard that defines 400GbE: the physical layer, management, and MAC parameters that switches and NICs must meet to interoperate.
OIF 400ZR specifies how to carry 400GbE traffic over DWDM fiber for data center interconnects, typically up to roughly 120 kilometers on amplified links.
OpenZR+ extends that reach further for metro and regional distances, supporting flexible bandwidths.
Together, these standards are what make it possible to buy a 400G transceiver from one vendor, plug it into a switch from another, and have it actually work.
What are 400G transceivers, and which form factor should you choose?
A transceiver is the pluggable module that converts electrical signals from a switch or NIC into optical signals for transmission over fiber, and back again at the other end. For 400G, three physical form factors dominate the market: OSFP, QSFP-DD, and QSFP112. They all deliver the same 400 gigabit throughput, but they differ enough in size, power handling, and compatibility that the choice affects your broader hardware strategy.
OSFP (Octal Small Form-factor Pluggable)
The largest of the three, with room for better heat dissipation, which matters because OSFP modules typically run higher power. It was developed alongside 400G itself rather than adapted from an earlier standard, so it isn't backward compatible with older QSFP equipment without an adapter. That tradeoff buys headroom: OSFP is the form factor built to scale toward 800G and beyond.
QSFP-DD (Quad Small Form-factor Pluggable, Double Density)
Smaller than OSFP and backward compatible with earlier QSFP28 and QSFP56 modules, which makes it a practical choice if you're upgrading an existing QSFP-based deployment incrementally rather than replacing everything at once.
QSFP112
The smallest footprint of the three, running four lanes at roughly 112 gigabits each rather than eight lanes at 50 gigabits. It's also backward compatible with QSFP equipment, but its smaller lane count makes it less suited to a future jump to 800G than QSFP-DD or OSFP.
Form factor | Lane configuration | Typical max power | Backward compatible | Best suited for |
OSFP | 8 x 50G | Higher (larger thermal envelope) | No (adaptor required) | New builds targeting 800G and beyond |
QSFP-DD | 8 x 50G | Moderate | Yes, with QSFP28/56 | Incremental upgrades from existing QSFP gear |
QSFP112 | 4 x 112G | Lower | Yes, with QSFP | Smaller-footprint deployments not targeting 800G |
Each form factor is governed by its own multi-source agreement, the specifications published by the OSFP MSA, QSFP-DD MSA, and QSFP112 MSA groups, which is what keeps modules from different manufacturers interchangeable within each form factor.
What optical standards and reach options do you need to know?
Once you've settled on a form factor, the next decision is reach: how far does the signal actually need to travel? This is where the standard naming (SR, DR, FR, LR, ER, ZR) starts to feel like alphabet soup, but each one maps to a specific distance and fiber type.
VR and SR family (multimode fiber, tens of meters): Built for short, in-rack or adjacent-rack connections where the transceivers are close together and cost matters more than reach.
DR and FR family (single-mode fiber, up to a few kilometers): Suited to cross-room or cross-building links within a single campus or data center site.
LR, ER, and ZR family (single-mode fiber, from 10 km to 80+ km): Built for campus, metro, and long-haul data center interconnect, with ZR variants using coherent optics and DWDM tuning to stretch even further.
Standard family | Typical reach | Fiber type | Common use case |
VR / SR | Up to ~100m | Multimode | In-rack and adjacent-rack |
DR / FR | Up to ~2km | Single-mode | Cross-room, cross-building |
LR / ER | 10km to 40km | Single-mode | Campus and metro |
ZR / ZR+ | 80km and beyond | Single-mode, DWDM tunable | Long-haul DCI |
The practical takeaway: match the standard to the actual distance you need, not the other way around. Over-specifying reach adds unnecessary cost and power draw, while under-specifying it means a link that simply won't close.
DAC, AOC, or transceiver and fiber: which cabling approach fits your rack?
Not every 400G link needs a transceiver and a separate fiber patch cable. For the shortest runs, two other options are usually more cost-effective.
Direct Attach Copper (DAC)
A cable with the transceiver electronics built into each end, using twinaxial copper to carry the signal with no optical conversion. DAC keeps cost and power draw low and adds essentially no latency, but its reach is limited, typically a few meters for passive variants, which makes it a natural fit for in-rack, switch-to-switch, or switch-to-NIC connections.
Active Optical Cable (AOC)
A similar concept to DAC, but with the optical transceiver built into the cable ends instead of copper conductors. AOC extends reach well beyond copper (commonly tens of meters) and avoids the electromagnetic interference issues copper cabling can run into in dense racks, at the cost of higher power draw from the electro-optical conversion.
Transceiver plus fiber patch cable
The standard approach once you're beyond the reach of DAC or AOC, using the form factors and optical standards covered above.
The rule of thumb: DAC for the shortest in-rack runs where cost and latency matter most, AOC for slightly longer runs where copper's reach or interference resistance becomes a problem, and transceivers with fiber patching for anything beyond that.
How do 400G switches and interconnects fit together in a real network?
Individually, transceivers, optical standards, and cabling are just components. What makes a 400G network work is how they connect switches, network interface cards, and each other into a coherent topology.
In most modern data center designs, that means a spine-leaf architecture: leaf switches connect directly to servers and storage, while spine switches interconnect the leaf layer. 400G links typically show up first at the spine-to-leaf and spine-to-spine layers, where aggregate bandwidth demand is highest.
Direct switch-to-NIC connections matter increasingly for AI and GPU cluster deployments, where east-west traffic between compute nodes can dwarf traditional north-south traffic patterns. A single failed or congested link in that kind of topology can stall an entire training run rather than affecting one workload, which puts a premium on getting the interconnect design right the first time.
Breakout configurations add flexibility during a migration. A single 400G switch port can be split into 2x200G or 4x100G links using a breakout DAC or AOC, letting you connect to existing 200G or 100G equipment without waiting for every device in the path to support native 400G.
What does migrating from 100G or 200G to 400G actually involve?
For most organizations, moving to 400G is a phased process that starts wherever the bandwidth pressure is highest, usually the spine layer or specific high-traffic interconnects, and expands from there.
The consolidation math is straightforward: a single 400G port does the work of four 100G ports, which means fewer optics, fewer cables, and less aggregate power draw for the same total capacity. Before committing to a migration path, a few practical questions are worth answering:
- Does your existing switch hardware support 400G natively, or will you need new chassis or line cards? This determines whether the migration is a software and optics update or a full hardware refresh.
- Which form factor makes sense given what's already deployed? QSFP-DD and QSFP112 backward compatibility can ease a phased upgrade in ways that OSFP, built purely for the 400G-and-beyond era, doesn't.)
Where can breakout configurations bridge the gap between existing 100G or 200G equipment and new 400G switches, rather than requiring a simultaneous upgrade across every device?
AddOn Networks works through this kind of data center scaling and upgrade planning with customers, including how compatible optics affect both capex and ongoing operating costs during a phased migration.
What comes after 400G?
400G won't be the ceiling for long. 800G deployments are already underway in the largest environments, using the same OSFP and QSFP-DD form factors covered above, often reached via breakout from existing 400G infrastructure rather than a wholesale replacement. 1.6T is next on the standards roadmap, driven by the same AI and GPU cluster growth that accelerated 400G adoption in the first place.
For a closer look at how AI workloads specifically are shaping this next jump, see AddOn's guide to transceivers and cabling built for AI infrastructure.
About AddOn Networks
AddOn Networks designs and manufactures fiber and copper connectivity solutions for data centers, enterprises, service providers, and governments. The company operates an in-environment network testing lab and holds ISO certification across its manufacturing and quality processes, backed by a support team built to help customers work through exactly the kind of form factor, standard, and migration decisions covered in this guide.
Ready to spec your 400G deployment?
Whether you're planning a first 400G rollout or a phased migration from existing 100G or 200G infrastructure, AddOn's team can help you match the right transceivers, cabling, and optics to your network. Explore AddOn's 400G transceivers and network cabling, use the transceiver finder and cable finder tools, or contact AddOn's support team directly.
What does 400G actually mean?
400G describes how much data a network connection can carry: 400 gigabits every second, or roughly four times the capacity of a standard 100G connection. It's less about any single piece of data moving faster and more about how much data can travel across that connection at once.
Why would a network need 400G instead of just staying on 100G?
Traffic keeps growing. AI training, cloud services, and video are all pushing more data through networks than they were built to handle a few years ago. Rather than adding more and more 100G connections and cables to keep up, many organizations move to 400G because a single 400G link carries roughly the same total traffic as four 100G links, with less equipment and less power required to do it.
What are the basic building blocks of a 400G network connection?
A working 400G connection generally needs three things: a transceiver (the module that turns electrical signals into light and back again), a cable or fiber connection to carry that light between devices, and a switch or network card at each end capable of handling 400G speeds. This guide walks through how each of those pieces works and how to choose between the available options.
Is 400G the same as 400GbE?
400G is the broader term for 400 gigabit capacity on a link or wavelength, while 400GbE (400 Gigabit Ethernet) is the specific IEEE 802.3bs standard governing how switches, routers, and NICs communicate at that speed.
Do I need to replace my existing fiber to move to 400G?
Often not, because many 400G standards, particularly the DR, FR, and LR families, run over the same single-mode fiber already common in most data centers. The transceivers, switches, and NICs are what change, not necessarily the fiber plant itself.
What's the practical difference between DAC and AOC for 400G links?
DAC uses copper and works best for the shortest, in-rack runs where cost and latency matter most. AOC uses fiber optics built into the cable assembly and extends reach further while avoiding copper's electromagnetic interference issues, at a higher power cost.
How much power does a 400G transceiver typically use?
It varies by form factor and reach. OSFP modules, built with a larger thermal envelope, generally run higher power than QSFP112, with QSFP-DD falling in between. Reach matters too: coherent ZR optics draw more power than short-reach multimode options.
What's the real difference between OSFP and QSFP-DD?
Size and compatibility. OSFP is larger, was purpose-built for 400G and beyond, and isn't backward compatible with older QSFP gear without an adapter. QSFP-DD is smaller and backward compatible with QSFP28 and QSFP56 modules, making it a more natural fit for incremental upgrades.