Upscale AI, a start-up based in Santa Clara, California, announced plans to deliver open, scale-out Ethernet systems built on NVIDIA Spectrum-X switch silicon and an AI-optimized SONiC software stack, targeting the next generation of heterogeneous AI data center clusters. The Santa Clara–based company said the new systems will enable operators to deploy low-latency, interoperable fabrics designed to connect GPUs, accelerators, storage, and memory across large-scale AI infrastructure.
The initiative aligns Upscale AI with the NVIDIA Partner Network and includes collaboration on reference architectures and validated designs based on the Spectrum-X Ethernet platform. The company positions its offering as a full-stack AI networking solution combining hardware, software, and lifecycle services. Upscale AI’s architecture focuses on enabling multi-vendor AI clusters while maintaining operational consistency through an enterprise-grade SONiC distribution tailored for AI workloads.
Upscale AI said its systems integrate ASIC-native telemetry, deterministic and lossless Ethernet behavior, and standard network operations workflows to support predictable performance at scale. The design supports high-speed data movement, workload isolation, and orchestration across heterogeneous compute environments. Upscale AI plans to bring the Spectrum-X–based scale-out systems to market later this year for hyperscalers, cloud providers, and AI infrastructure operators deploying large-scale clusters.
- Upscale AI plans open Ethernet scale-out systems built on NVIDIA Spectrum-X switch silicon
- Systems use an AI-optimized SONiC network operating system with full lifecycle support
- Platform targets heterogeneous AI clusters spanning GPUs, accelerators, storage, and memory
- Architecture integrates ASIC telemetry and deterministic lossless Ethernet behavior
- Solutions designed for multi-vendor AI infrastructure and open operational workflows
- Commercial availability expected later in 2026
“Pairing NVIDIA Spectrum-X Ethernet switch silicon with our purpose-built systems and AI-optimized SONiC software allows us to deliver an open, highly scalable and interoperable architecture with operational simplicity,” said Barun Kar, CEO of Upscale AI.
🌐 Analysis
NVIDIA introduced Spectrum-X as an Ethernet networking platform designed specifically for AI workloads. The architecture combines Spectrum Ethernet switch ASICs with adaptive routing, congestion control, and telemetry capabilities tuned for distributed GPU training and inference clusters. The platform also integrates tightly with NVIDIA’s BlueField DPUs and AI Enterprise software stack to optimize traffic flows between GPUs across large-scale clusters. NVIDIA positions Spectrum-X as a high-performance Ethernet alternative for scale-out AI networking, aiming to deliver deterministic performance and improved link utilization compared with conventional data center Ethernet fabrics.
SONiC (Software for Open Networking in the Cloud) is an open-source network operating system originally developed by Microsoft for hyperscale cloud data centers. Built on a modular Linux-based architecture, SONiC separates control-plane functions from the underlying switching hardware through standardized abstraction layers, enabling operators to run the same software stack across multiple switch vendors. The platform has gained wide adoption among hyperscalers and enterprise operators seeking programmable, open networking environments. AI-focused implementations often integrate additional telemetry, congestion management, and automation capabilities to support the high-bandwidth, low-latency requirements of large-scale AI clusters.
🌐 We’re tracking the latest developments in AI infrastructure and data center networking. Follow our ongoing coverage at: https://convergedigest.com/category/data-center/