Scaling Inference Lab

The Scaling Inference Lab

The Scaling Inference Lab is a testbed for AI hardware technologies, prioritising rapid iteration, open collaboration, and long-term sustainability.

Backed by £50m, the Lab is set up to address a key bottleneck for startups and research groups developing innovative AI hardware: the lack of an open platform to showcase their technologies due to the dominance of large, closed systems in today’s AI hardware market.

Delivered by CommonAI CIC, the Lab will create real and deployable rack-level Al systems, designed to increase the ability of startups to insert new technologies with the goal to rapidly reduce cost of large AI systems. A new rack will be built every 6 months and serve as a 'structured pilot', incorporating one unproven technology (accelerators, memory, interconnects, etc.), and bringing it up in a system such that actual end users can run Al models on it. For each build, we should be able to compare the full systems-level performance of a particular technology to existing commercial offerings.

A photo of Suraj Bramhavar in front of a blue curtain.

Inaugural partners

Partners in the Lab provide services, technology, or expertise either for individual clusters or across the full initiative.

CommonAI

CommonAI CIC will lead the delivery of the Scaling Inference Lab, both operationally and through a dedicated research and engineering effort. The CommonAI model and its key contributors have a proven track record of moving ideas from open-source research to global scale. This was exemplified by the openTitan project which was incubated by the lowRISC CIC and is being deployed globally by Google and others.

Website | LinkedIn

Callosum

Callosum's technology orchestrates diverse AI systems across heterogeneous chips, enabling multi-agent inference architectures that no single-chip platform can support. As part of the Lab, Callosum will establish a new inference scaling law, demonstrating that heterogeneous orchestration delivers dramatically better cost-performance gains than single-model approaches.

Website | LinkedIn

Sequrity.AI

Sequrity.AI are reimagining sustainable hardware solutions for AI workloads to ensure a secure and reliable end-to-end inference supply chain. As part of the Lab, they will provide software solutions to optimise our initial clusters, and help ensure these clusters are benchmarked against the best commercial offerings.

Website | LinkedIn

Oriole Networks

We are in advanced talks to partner with Oriole, who are pioneering a new class of AI infrastructure powered entirely by photonics, replacing today’s electronic bottlenecks with nanosecond‑speed optical circuits. Within the Lab, Oriole will stand up an AI cluster on the PRISM fabric to validate performance and integration. This pilot will be the foundation for a radically new architecture in which the entire data centre functions as a single, tightly coupled accelerator, unconstrained by topology or switching limits.

Website | LinkedIn

Become a partner

We are now inviting expressions of interest from organisations who would like to participate in the development of future racks.

Each cluster will be built in partnership with at least one external organisation, with ARIA funding system integration and testing. Participating partners will be expected to provide matched technical and/or financial contributions for technology development.

If you would like to explore participation, express your interest by completing the form below. The selection process will be run by CommonAI CIC.

Support from the ecosystem

“The Scaling Inference Lab creates a practical environment where new AI infrastructure can be tested and proven at system scale. It builds on CommonAI's vision of shared infrastructure, allowing organisations to innovate without needing the scale or resources of large technology providers. By improving access to efficient, trusted computing platforms, we can help create a more accessible AI ecosystem and unlock greater economic opportunity across the UK.”

Sir Andy Hopper

Chairman, CommonAI CIC

"AMD believes that research with open testbeds is essential to expanding the capabilities of machine learning. Open, collaborative environments help accelerate experimentation and translate breakthroughs into real-world impact. We are excited about the potential of ARIA’s Scaling Inference Lab to enable the next generation of inference innovation and help advance more scalable, efficient AI deployments."

Ralph Wittig

Corporate Fellow, CVP Head of Research & Advanced Development, AMD

"The future of AI infrastructure cannot be built on closed, vertically integrated stacks. In collaboration with ARIA, we are advancing an open AI rack architecture based on Open Chiplet Atlas ™ (OCA) that enables true multi-vendor composability. Open interfaces at every layer will unlock innovation, reduce systemic risk, and fundamentally reshape the cost curve of AI systems.”

Wei-Han Lien

Chief Architect and Senior Fellow at Tenstorrent

"To lead in AI, the UK must lead in the silicon and systems that power it. ARIA’s Scaling Inference Lab is exactly the catalyst we need to break open the hardware ecosystem, drive innovation, and ensure the UK is building the physical foundation of the AI revolution.”

Nick McKeown

Professor Emeritus, Stanford University and Visiting Professor of Engineering, University of Oxford

Sir Andy Hopper

Chairman, CommonAI CIC

Ralph Wittig

Corporate Fellow, CVP Head of Research & Advanced Development, AMD