What is scale up, scale out and scale across?
Three ways to scale compute capacity
Scale up, scale out and scale across describe three ways to increase compute capacity, performance and reach. To scale up is to make a single system more powerful by adding resources such as CPU, memory, GPUs or bandwidth. To scale out is to add more systems or nodes and distribute workloads across them. In AI infrastructure, scale across extends this further by connecting clusters, data centers or cloud environments so workloads can run across distributed sites. Together, these approaches help organizations support larger applications, heavier data movement and increasingly demanding AI workloads.
Scale up
Scale up increases the power of a single system to deliver higher performance with minimal latency and architectural complexity.
Effective scaling balances performance, capacity and reach by combining scale-up, scale-out and scale-across approaches.
Scale out
Scale out increases overall capacity by adding systems and distributing workloads across multiple nodes within a larger cluster.
How scaling works in AI infrastructure
AI workloads typically rely on all three approaches working together. Scale-up architectures support tightly coupled compute environments where accelerators require extremely fast, low-latency communication within a server, rack or small group of racks. Scale-out architectures extend this across clusters, allowing multiple systems to process workloads in parallel. Scale-across architectures then connect clusters, data centers or cloud platforms, enabling distributed infrastructure to operate as part of a broader compute fabric. As demand grows, network performance becomes critical, since data must move quickly and reliably between systems.
Scale across
Scale across connects clusters, data centers or cloud environments, enabling distributed infrastructure to operate as a unified, large-scale compute system.
Supporting AI scaling with Adtran
As AI workloads grow, the network must scale with them. High-performance environments require not only more compute but also low-latency, high-capacity connectivity across systems, sites and services. Adtran supports this evolution with scalable optical transport, data center interconnect and cloud networking technologies that enable efficient data movement across distributed infrastructure. Combined with intelligent network operations and service assurance, these capabilities help service providers and enterprises build networks that remain resilient and efficient as workloads scale up, scale out and scale across.
;