Fibre optics for the next capacity
From traditional server environments to NVIDIA-based AI clusters. We help you choose the right optics, cabling and connectivity for 100G, 400G and 800G.
1
1G-800G
Optics and connections
2
25,000+
Products in the range
3
Fast delivery
From stock to deployment
4
40+ years
Combined experience and expertise
NVIDIA-based AI environments
The right connections for AI infrastructure
In an NVIDIA-based AI cluster, many GPU servers work in parallel. Large amounts of data need to be moved quickly between servers, network cards, and switches. This places high demands on bandwidth, low latency, and stable performance in each connection.
GPU servers
Parallel AI workloads
SuperNIC
Connects the server to the network
Specturm-X
/InfiniBand
Parallel AI workloads
Tellus Fiber connects the infrastructure
- DAC and AOC for short connections between servers and switches
- 400G and 800G transceivers for high-capacity optical links
- MPO/MTP and fibre cabling between racks and halls
- Breakout when a high-capacity port is to be split into multiple links
Selected products for datacenters
What changes?
When more traffic stays in the data center
AI, virtualisation, and distributed storage mean that an increasing share of traffic goes between servers, switches, and compute clusters.
Traditionally
A lot of traffic goes between users, applications, and the internet.
Cloud
Servers synchronise data and distribute workloads across more nodes.
AI cluster
GPUs continuously exchange large amounts of data. The network's capacity directly affects how efficiently the cluster operates.
The right solution for every link
Direct paths to your most important product categories - from server connection to data centre interconnect.
100-800G
QSFP8, QSFP-DD and OSFP for backbone, spine-leaf, AI and HPC.
DAC and AOC
Cost-effective connections with high capacity within and between racks.
Breakout
Split 400G and 800G into multiple links and use each port efficiently.
MPO/MPT
High-density cabling for a structured and scalable infrastructure.
CWDM and DWDM
Increase the capacity of existing fibre between halls and facilities.
Fibre cabling
Custom lengths, fibre types and connectors for your installation.
From requirements to functioning link
We help you check speed, range, fibre type, connectors, breakout, FCE and coding before ordering.
When servers talk to servers
A large part of the traffic in a datacentre is not out to the internet – it is internal. Servers that sync databases, share computing power, or store the same files in multiple locations. This communication, often called east–west traffic, has exploded alongside virtualisation, cloud services, and AI, where NVIDIA is dominant.
Over short distances, multimode and SR transceivers are sufficient – perfect for connecting between servers and to nearby switches. But when traffic needs to pass between entire server clusters or different halls, singlemode and transceivers such as LR or ER are required. Each link becomes a vein in the system, and stability is absolutely crucial.
From 400G to what you actually need
Datacentres are built with ports that can handle 100G, 400G, 800G and up to 1.6TB – but rarely does a single server need all that space. Therefore, the capacity can be divided. For example, a 400G port can be transformed into four separate 100G links, ready to feed different servers. Similarly, multiple signals can be combined on a single fibre, or broken out again at the other end.
It's about utilising resources smartly – ensuring that every port, every fibre, every transceiver is used to the maximum.