Upscale AI & NVIDIA: Scaling Heterogenous AI Infrastructure

As companies consume immense AI power, Upscale AI has announced its plans to supercharge open and heterogeneous scaling out of AI clusters. 

Ethernet systems are the bedrock of this scale up, and will be built on industry reputed NVIDIA Spectrum X Ethernet switch silicon, along with Upscale AI’s SONiC software that is optimised for AI. 

The aim is to deploy scalable, low-latency fabrics, which are optimised for today’s intensive AI workloads and are spread across heterogeneous architectures in compute, accelerators, memory and storage.

“NVIDIA Spectrum-X Ethernet switch silicon is setting a new standard for Ethernet-based AI performance," says Barun Kar, CEO of Upscale AI.

“Pairing this technology with our purpose-built systems and AI-optimised SONiC software allows us to deliver the best of both worlds: an open, highly scalable and interoperable architecture with operational simplicity.”

What is the NVIDIA Partner Network? 

The NVIDIA Partner Network is a globally renowned multi-tier partner ecosystem, where members deploying NVIDIA platforms and architecture can gain specialised expertise and grow their revenue by accessing NVIDIA’s validated, cutting-edge AI, data centre and visualisation technologies.

Strengthening its position as a “pureplay provider of NVIDIA AI-native networking infrastructure”, Upscale AI will be joining the NVIDIA Partner Network. 

This offers Upscale AI the opportunity to work within the NVIDIA ecosystem, wherein reference architectures and validated designs are available, bolstering Upscale's ambition to scale AI data centre networks.

Open AI Infrastructure at Scale 

To ease complexity for heterogeneous AI clusters to work together, Upscale AI supports open and interoperable ethernet networking. 

The AI fabric of Upscale is designed by taking into account the diversity of compute environments, deployed using multi-vendor, production-grade AI infrastructure, while seamlessly operating on top of a consistent operating model. 

AI-optimised scale out systems, built over NVIDIA hardware, also integrate ASIC-native telemetry – where the chips directly generate, collect and export real-time performance data – thereby offering lossless Ethernet behaviour.

Predictable performance, operational simplicity and reliability at scale are other advantages, along with end-to-end support offered by SONiC’s enterprise grade network operating system.

“To lead in the trillion-parameter model era, scalability and efficiency are paramount,” says Gilad Shainer, SVP at NVIDIA. 

“We look forward to collaborating closely with Upscale AI as the team leverages the NVIDIA Spectrum-X Ethernet platform to help companies build the world's most advanced open AI infrastructure.”

Practical scale up for heterogeneous environments

Upscale AI offers a full stack approach that paves way for high-speed data movement, workload isolation and scalable orchestration, in environments where the system contains different types of compute architectures or technologies.

Combining hardware, software and lifecycle services, Upscale AI’s Spectrum X Ethernet scale-out systems also preserves the flexibility of open-source management.

“As AI infrastructure evolves toward increasingly heterogeneous architectures, scalable and operationally sound Ethernet fabrics are becoming essential," says Alan Weckel, Co-Founder and Technology Analyst at 650 Group. 

“By combining NVIDIA Spectrum-X Ethernet switch silicon with a full-stack approach that integrates systems, software and support, Upscale AI is addressing a key gap in making open, scale-out AI networking practical for large scale AI deployments.”

These scale out systems will hit the market later in 2026 available for AI data centres building diverse, multi-vendor infrastructure.