Endpoint density
More GPU, server and storage ports increase link counts, optics at both ends and spare requirements.
Why requirements are changing
More endpoints, faster fabrics and denser racks change the physical-link plan.
Modern infrastructure can combine GPU clusters, servers, storage, campus systems, cloud interconnects and provider networks. The optical plan must account for every port, both ends of each link and the physical conditions in which the connection will operate.
100G to 800G
Higher speeds introduce new lane arrangements, breakout choices, connector formats and power considerations.
Mixed link domains
One environment may combine inside-rack, leaf–spine, campus, metro and long-reach connections.
Physical constraints
Rack layout, airflow, bend radius, cable routing and installed fiber can change the suitable medium.
Platform qualification
Port capability, software, coding and supported operating modes must be confirmed before selection.
Migration choices
Fiber, connector and patching decisions made today can affect a later move to higher port speeds.
“AI factories are a new class of data centers with extreme scale, and networking infrastructure must be reinvented to keep pace.”Jensen Huang — Founder and CEO, NVIDIAREAD
“Optical interconnect is no longer a luxury; it is a necessity for hyperscale AI to continue its exponential growth path.”Dr. Wei-Chung Lo — Deputy General Director, EOSL, ITRI TaiwanREAD
A better starting point
Define the requirement through three independent dimensions.
Cloud model, workload and physical architecture are related, but one does not automatically determine the others. An AI workload, for example, may operate in a neocloud, public cloud, sovereign cloud or private enterprise environment.
01
Where will it operate?
Operating model
02
What must it support?
Workload and purpose
03
What is being connected?
Physical architecture
Optical connectivity scenarios
Move from the planning context to the physical connections.
Explore practical link patterns across four network environments. The scenarios help structure connectivity inputs; they are not complete network or cloud reference designs.
AI & Data Center
GPU and server downlinks, inside-rack links, leaf–spine fabrics, breakout connectivity and data-center interconnect.
Explore AI scenarios →Enterprise & Campus
Fiber access links, distribution and core uplinks, building backbones and private-cloud connectivity.
Explore campus scenarios →WAN & Router
Router-to-router, branch, cloud on-ramp, metro and long-reach links built around the handoff and path.
Explore WAN scenarios →Service Provider & Telecom
Access, aggregation, provider edge, metro transport, telco-cloud and mobile-infrastructure connectivity.
Explore provider scenarios →Count the complete connection
Identify every link domain before selecting the product.
A useful BOM distinguishes endpoint connections from network uplinks and counts the required termination at both ends.
- GPU, server and storage downlinks
- Switch uplinks and leaf–spine connections
- Native and breakout topologies
- Inter-building, WAN, metro and long-reach links
- Optics at both ends, cable assemblies and spares
- Out-of-band management and console requirements
Choose the connectivity medium
Match the construction to the physical link.
Distance alone is not enough. Port support, lane topology, cable routing, connector strategy, serviceability and installed fiber all influence the decision.
DAC
Best suited for Short, fixed copper connections within or between nearby racks.
Primary consideration Low power and cost, balanced against bulk, bend radius and supported reach.
Explore DACs →AOC
Best suited for Lightweight, factory-terminated short-to-medium cable runs.
Primary consideration Simple deployment where fixed optical ends suit the cabling plan.
Explore AOCs →Pluggable optics
Best suited for Structured fiber, flexible reach and independently replaceable modules.
Primary consideration Correct transceiver, connector, fiber type and optical budget.
Explore transceivers →Breakout assemblies
Best suited for Dividing one higher-speed host port into multiple lower-speed endpoints.
Primary consideration Parent port, lane mode, endpoint count, endpoint form factor and length.
Explore breakouts →Fiber assemblies
Best suited for Patch panels, cross-connects and structured optical cabling.
Primary consideration Connector, polarity, fiber type, route and patching design.
Explore fiber assemblies →How NRCube can help
Translate a defined architecture into clearer optical requirements.
NRCube does not design the customer’s cloud, data center or network architecture. We work from requirements supplied by the customer or system integrator and help organize the corresponding optical products and quantities.
About NRCube- Review platform, port and operating-mode information
- Map documented industry PIDs to NRCube products
- Compare technically suitable optical options
- Review speed, form factor, connector, fiber and reach
- Reconcile links with required terminations at both ends
- Identify missing or conflicting BOM information
- Prepare a proposed optical-connectivity BOM
- Review alternatives based on availability and lead time
Complimentary pre-quotation service
AI-assisted requirement review
Customer BOMs and connectivity assumptions can be checked against structured NRCube product data to highlight missing inputs, quantity inconsistencies, possible PID mappings, alternative products and assumptions requiring confirmation.
AI supports the analysis. NRCube reviews the proposed output before it is shared. The review does not replace platform approval, architecture design, deployment or fault isolation.
01Customer BOMPorts, quantities, reach and notes
02Structured reviewGaps, mappings and alternatives
03NRCube checkAssumptions and clarifications
04Proposed BOMProducts and reviewed quantities
From requirement to review
Start with the architecture you already have.
Identify the operating context, workload and physical architecture; enter the link assumptions; then request NRCube review of the preliminary connectivity estimate.
Context+Workload+Architecture→Link estimate→NRCube review