Upscale AI & NVIDIA: Scaling AI Data Centre Networks

As organisations demand ever greater computing power for AI workloads, the pressure on data centre infrastructure is growing. Clusters that run advanced AI models require both high-speed networking and the ability to scale across diverse hardware environments.

Upscale AI has set out plans to strengthen this foundation through open and heterogeneous scaling for AI clusters, building its systems on NVIDIA Spectrum-X Ethernet switch silicon. It combines this hardware with its own SONiC-based software platform designed specifically for AI data centres.

Ethernet networking forms the core of this strategy, as Barun Kar, CEO of Upscale AI explains: “NVIDIA Spectrum-X Ethernet switch silicon is setting a new standard for Ethernet-based AI performance.

“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.”

Joining the NVIDIA ecosystem

Upscale AI strengthens its position as what it describes as a pureplay provider of NVIDIA AI-native networking infrastructure. As part of this approach, the company is joining the NVIDIA Partner Network.

The NVIDIA Partner Network is a global partner ecosystem built around NVIDIA platforms and architecture. The programme provides multiple tiers of partnership and allows companies to develop specialised expertise using NVIDIA technology.

Access to NVIDIA’s data centre and visualisation technologies also supports the development of high-performance computing environments. Through the programme, partners collaborate with NVIDIA and expand their capabilities around accelerated computing.

Upscale AI's membership in the ecosystem aligns with its aim to scale AI networking within modern data centres. The collaboration allows the company to design infrastructure around NVIDIA Spectrum-X Ethernet technology while working within a broader ecosystem of partners that build compatible systems.

Open infrastructure for large AI clusters

AI clusters within data centres increasingly rely on heterogeneous architecture, meaning systems that contain different types of processors and hardware, including CPUs, graphics processing units, specialised AI accelerators, memory platforms and storage technologies.

These environments require networking infrastructure capable of connecting multiple technologies while maintaining consistent performance across the cluster. Upscale AI approaches this challenge with open and interoperable Ethernet networking.

The company’s AI fabric accounts for the diversity of these compute environments. Systems deploy using multi-vendor production-grade infrastructure while operating through a consistent operating model that simplifies management.

Upscale AI’s scale-out systems, built on NVIDIA hardware, also integrates ASIC-native telemetry. An ASIC, or application-specific integrated circuit, is a chip designed for a particular computing function.

With ASIC-native telemetry, the networking chips generate, collect and export real-time performance information directly from the hardware. This provides visibility into network conditions and helps maintain lossless Ethernet behaviour, which is designed to prevent packet loss during heavy traffic loads.

In large AI clusters where thousands of processors exchange data during model training, preventing packet loss and maintaining low latency communication are critical to sustaining performance.

“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.”

Building practical scale-out data centre systems

Upscale AI adopts a full-stack strategy for its networking platforms. For data centre operators, this model supports high-speed data movement across AI clusters while maintaining workload isolation.

Upscale AI’s Spectrum-X Ethernet scale-out systems combine networking hardware with AI-optimised software and lifecycle services.

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

“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.”

The platform maintains compatibility with open-source management frameworks, which allows operators to retain flexibility when managing multivendor infrastructure inside data centres.

Upscale AI expects its Ethernet scale-out systems to enter the market later in 2026.