Scale-beyond

The connectivity fabric for distributed AI compute

Scale-beyond

What is scale-beyond?

Scale-beyond is the connectivity fabric for geographically distributed compute clusters facilitating the federation of independent AI compute resources. Scale-beyond networks are optimized to provide the performance, security, resiliency, and scalability necessary to enable the coordination and distribution of workloads as well as the necessary end-user connectivity seamlessly in coordination with a higher-level workload orchestrator.

Network scaling for AI infrastructures

What is driving the distribution of compute resources for AI?

As AI solutions move from generative and agentic into physical real-world applications, inferencing optimized compute resources will be increasingly geographically distributed. This includes bringing resources closer to the end user which is critical to enable the performance, cost, latency, and sovereignty necessary to support many of the most important applications.

Blog

Scale-beyond: when AI systems become more than their clusters

What are the essential characteristics of a scale-beyond network?

Scalable

Elastically allocate and optimise networking capacity to meet changing AI compute demands, ensuring efficient resource utilization and cost-effective service delivery.

Dedicated

Provide isolated and assured network resources tailored to the performance, reliability, and security requirements of distributed AI workloads.

Programmable

Provide an open and automated network where scale, capacity, and performance are coordinated in conjunction with an AI workload orchestrator.

Data center gateway and WAN routing solutions

Carrier-grade routing solutions designed to handle advanced and secure networking functions for interconnecting AI resources. Support a range of solutions to meet the diverse requirements of an AI grid.

Learn more

Efficient optical connectivity

Build using the latest optical technologies and manufactured in our own US-based fab, our solutions are cost, space, and power optimized to facilitate efficient network connectivity at any capacity and across any distance.

Learn more

Why is scale-beyond connectivity needed?

Data centers currently leverage several connectivity techniques to increase the size of AI clusters enabling connections between hundreds of thousands or even millions of AI processors. These techniques include scale-up which connects multiple processors across one or two racks, scale-out which connects multiple racks of processors within a single physical location, and scale-across which connects racks of processors across multiple physical locations. However, these solutions only provide connectivity for a single logical AI compute cluster.

Scale-beyond is a solution designed to connect a large number of widely geographically distributed compute clusters where each contains their own compute resources. These networks are optimized to provide the performance, security, resiliency, and scalability necessary to enable the coordination and distribution of workloads as well as the necessary end-user connectivity seamlessly in coordination with a higher-level workload orchestrator.

Various dimensions of scale in AI networks

Customer success stories

Customer success

Nokia and Nscale partner to accelerate AI infrastructure build-out

Customer success story

Gcore’s AI Cloud powered by Nokia data center networks

Customer success

Nokia powers Telefónica's AI-ready edge data centers in Spain

Customer success

Nokia selected by CoreWeave to provide networking backbone behind Ai and cloud infrastructure

Frequently asked questions

What is distributed inferencing and how does it relate to scale-beyond?

Distributed inferencing extends the reach of centralized AI factories into distributed compute environments. By federating these environments across cloud, regional, edge, and access into a single cohesive operational infrastructure, it becomes the AI infrastructure framework through which intelligence can be delivered everywhere it is needed. Scale-beyond is the connectivity layer that connects all of these locations.

Who will deploy a scale-beyond network?

A scale-beyond network can be deployed in any network that requires connectivity for distributed compute. This network can support AI infrastructures from hyperscalers, telcos, neoclouds, and large enterprises.

For a scale-beyond network do you need dedicated routers and optical transport?

You will need routing and optical transport resources that operate in an isolated and non shared manner. This does not always mean a requirement for a new set of routers and optical switches. It does require secure isolation of resources so there is no contention within this specific network.