The race to support higher network speeds shows no signs of slowing down. As traffic patterns shift from on-premises computing, to the cloud, and now to AI workloads, network infrastructure is under constant pressure to scale.
Copper cabling systems, while useful for short runs, are increasingly limited in their ability to cost-effectively support meaningful distances at higher speeds. This has shifted the responsibility to optical fibre systems, which can deliver the performance and scalability modern applications demand.
Contents
Parallel Optics and Connector Evolution
To achieve higher speeds, networks are moving toward parallel optics. This shift has also transformed connector technology:
- From simple two-fibre duplex connectors,
- To multi-fibre connectors,
- And now to Very Small Form Factor (VSFF) connectors.
Each step reflects the growing need for compact, high-density solutions that maintain signal integrity at higher data rates.
The Role of Standards: IEEE, TIA, and TSB 6000
Standards organizations have been central in guiding this evolution. IEEE and TIA define network speeds,
supported distances, and connector requirements. As technologies mature, Telecommunications Systems Bulletins (TSBs) provide interim guidance until full standards are updated.
In January 2025, TIA released TSB-6000, a key reference that consolidates:
- Channel attenuation guidelines
- Supported distances for Ethernet, InfiniBand, and other interfaces
- Media options across copper, fibre, and coaxial cabling
For designers, TSB-6000 offers a quick way to compare supported distances on multimode and
single-mode fibres across applications.
Ethernet vs. InfiniBand
- InfiniBand is designed for high-performance computing (HPC), offering very high throughput and ultra-low latency. It is widely used in hyperscale data centres.
- Ethernet (IEEE 802.3) remains the most common networking technology in data centres and enterprise networks.
Interestingly, Ethernet adoption is increasing even in AI networks. As InfiniBand adopts Forward Error Correction (FEC) to maintain performance, its latency advantage over Ethernet is narrowing, making Ethernet a stronger contender.
This shift is especially visible in modern data centres, where high-density optical connectivity and scalable cabling systems form the backbone of AI-ready infrastructure.
Interfaces and Transceiver Form Factors
Network applications rely heavily on optical transceivers and active optical cables (AOCs).
- AOCs: Fibre bonded with a transceiver; commonly used in InfiniBand.
- Transceivers: Modular (pluggable), with electrical input on one end and optical output on the other. More common in Ethernet structured cabling systems.
Common Transceiver Types
| Form Factor | Lanes per Direction | Common Speeds | Typical Use |
|---|---|---|---|
| Small Form Factor Pluggable (SFP) | |||
| SFP / SFP+ | 1 (sometimes 2) | 1–10 Gb/s | Ethernet (short to medium distances) |
| SFP28 | 1 | 25 Gb/s | Ethernet (25G links) |
| SFP-DD | 2 (double density) | 50 Gb/s | Higher-density Ethernet |
| Quad SFP (QSFP) | |||
| QSFP+ | 4 | 40 Gb/s | Early 40G Ethernet |
| QSFP28 | 4 | 100 Gb/s | 100GbE, InfiniBand EDR |
| QSFP56 | 4 | 200 Gb/s | InfiniBand HDR, 200/400GbE |
| QSFP-DD | 4 | 200–400 Gb/s | High-density 200/400GbE |
| Octal SFP (OSFP) | |||
| OSFP | 8 | 200–800 Gb/s | Hyperscale, AI/ML workloads |
IEEE Standards – 100G, 200G and 400G
| Standard | Medium | Max Distance | Fibre Count |
|---|---|---|---|
| 100GBASE-CR4 | DAC (copper) | ~7 m | 4 pairs |
| 100GBASE-SR4 | OM4 MMF | ~100 m | 8 fibres |
| 100GBASE-LR4 | SMF (WDM) | 10 km | 2 fibres |
| 100GBASE-ER4 | SMF (WDM) | 40 km | 2 fibres |
| 200GBASE-CR4 | DAC (copper) | ~3 m | 4 pairs |
| 200GBASE-SR4 | OM4 MMF | 100 m | 8 fibres |
| 200GBASE-DR4 | SMF | 500 m | 8 fibres |
| 200GBASE-FR4 | SMF (CWDM) | 2 km | 2 fibres |
| 200GBASE-LR4 | SMF (WDM) | 10 km | 2 fibres |
| 400GBASE-DR4 | SMF | 500 m | 8 fibres |
| 400GBASE-FR4 | SMF (WDM) | 2 km | 2 fibres |
| 400GBASE-FR8 | SMF (WDM) | 2 km | 2 fibres |
Connectors: From Duplex LC to MPO and Beyond
Connector technology continues to evolve alongside transceivers:
- LC duplex connectors: common with SFP modules
- MPO (Multi-Fibre Push-On) connectors: common with QSFP modules, supporting rows of 12 fibres each
- SM MPO: APC polish
- MM MPO: UPC polish
As speeds climb past 200 Gb/s, low insertion loss, tight alignment, and signal integrity become critical. Vendors are increasingly specifying APC-polished multimode connectors to manage return loss and maintain performance under tighter link budgets.
Future-Ready Data Centres
As networks evolve toward 200, 400, and even 800 Gb/s, the data centre is where these technologies truly come together. High-density transceivers, advanced connector systems, and optimized fiber management are the building blocks of scalable, AI-ready infrastructure.
Explore how we are enabling next-generation Data Centre Solutions with high-density optical products designed for performance, reliability, and scale.
