Build AI Infrastructure Without Network Bottlenecks.
As AI deployments expand, connectivity requirements change across GPU domains, cluster fabrics, distributed infrastructure, and GPU availability. Explore how iPronics ONE addresses these different AI networking requirements.
Scale-Up Demands a New Class of OCS
As scale-up architectures grow, the requirement changes: more optical connectivity must fit within increasingly constrained rack space. iPronics ONE uses silicon photonics to deliver a compact, high-density OCS platform for this new scale-up challenge.
OCS Adoption Is Accelerating
Optical switching is expanding across AI data-center connectivity applications.
Scale-Up Increases Connectivity Demand
Larger GPU domains require significantly more optical connectivity.
Rack Space Becomes the Constraint
More optical switching must fit within limited rack space.
Silicon Photonics Enables a Compact Form Factor
Miniaturized photonic switching enables high port density within constrained rack space.
Scaling in Every Direction
Scaling GPU Domains with Optical Connectivity
As GPU domains grow, fixed connectivity becomes a scaling constraint. Configurable optical circuits allow physical connectivity to adapt as accelerator domains and workloads evolve.
01
Training at Scale
Provide high-bandwidth connectivity across large GPU domains for communication-intensive training.
02
Inference Connectivity
Provide low-latency optical paths for demanding inference workloads.
03
GPU Domain Expansion
Extend and reconfigure connectivity as GPU domains grow within and across racks.
Outcome
Scale larger GPU domains without locking infrastructure into fixed physical connectivity.
Expand AI Cluster Fabrics
As scale-out fabrics grow, more endpoints and traffic increase connectivity demands across the cluster. Configurable optical paths provide additional flexibility as cluster fabrics expand.
01
Fabric Expansion
Add optical paths as scale-out clusters and endpoints grow.
02
High-Volume Flow Placement
Establish paths for sustained, high-bandwidth east-west traffic.
03
Topology Reconfiguration
Scale larger AI cluster fabrics with greater flexibility and less dependence on fixed physical topology.
Outcome
Scale larger AI cluster fabrics with greater flexibility and less dependence on fixed physical topology.
Connect Distributed AI Infrastructure
As AI infrastructure expands across clusters, sites, and data centers, connectivity requirements shift with workload and data-movement demands. Configurable optical paths adapt across distributed infrastructure.
01
Multi-Site Training
Support high-bandwidth connectivity for distributed training across sites and data centers.
02
Distributed Inference
Support configurable connectivity for inference services across distributed deployment domains.
03
Bulk Data Movement & Replication
Provision optical paths for high-volume transfer, replication, and checkpointing.
Outcome
Move bandwidth where it is needed across distributed AI infrastructure without relying on fixed physical connectivity.
Connect Available GPU Capacity
When a GPU becomes unavailable, iPronics ONE can reconfigure optical connectivity to connect available GPU capacity.
01
GPU Failure
Redirect connectivity to an available GPU.
02
Planned Maintenance
Connect reserve GPU capacity during maintenance windows.
03
Reserve Capacity
Bring reserve GPU resources into the active infrastructure when needed.
Outcome
Reduce workload disruption by reconfiguring connectivity to available GPU capacity when accelerator resources become unavailable.