The development of artificial intelligence infrastructure is increasingly dominated by established computing architectures, making it difficult for smaller technology companies to demonstrate alternatives at meaningful scale. Testing new approaches to processing, networking and system integration requires access to expensive hardware and complete computing environments that few startups can afford independently.
A new UK research initiative is intended to address that barrier by providing shared infrastructure where emerging technologies can be evaluated against commercially available systems.
Kao Data has been selected by CommonAI to host the Scaling Inference Lab at its Harlow data centre campus, establishing a national testbed for experimental AI computing infrastructure. The project is funded by the Advanced Research and Invention Agency (ARIA) and delivered by CommonAI.
The agreement will see at least five high-density, GPU-accelerated clusters installed across two dedicated technology suites. Each will examine a different approach to computing infrastructure, including pipeline parallelism, heterogeneous orchestration and nanosecond optical circuit switching.
All five clusters are expected to become operational by early 2027, providing startups and researchers with access to complete rack-scale systems rather than isolated computing components.
The infrastructure will incorporate a mixture of NVIDIA and AMD technologies, including NVIDIA HGX B300 systems based on Blackwell Ultra, H200 NVL and RTX PRO 6000 Blackwell GPUs, alongside AMD Instinct MI350X accelerators. Servers will be supplied by Supermicro, MSI, Fujitsu and Dell.
This diversity is important to the lab’s purpose. By incorporating different computing architectures, the facility can support comparative testing of technologies intended to improve how AI workloads are distributed, processed and connected.
The project also introduces a different model for infrastructure experimentation. CommonAI plans to develop experimental clusters on a six-month cycle, allowing emerging approaches to be evaluated as computing requirements evolve.
ARIA committed £50 million to the Scaling Inference Lab earlier this year, while the UK Government announced more than £20 million in additional funding in June to establish a national capability for developing new AI computing systems.
Kao Data’s Harlow location places the facility between London and Cambridge, providing access to two important centres of UK technology research and startup activity. The operator says the infrastructure will be powered by 100 per cent certified renewable energy.
David Stringer, Partnerships Lead for the Scaling Inference Lab at CommonAI, said the objective was to provide innovators with a neutral environment in which to demonstrate technologies at rack scale.
For the data centre industry, the initiative illustrates how specialist facilities can support infrastructure development beyond conventional commercial hosting. The challenge is not simply providing additional computing capacity, but creating environments where alternative architectures can be tested under realistic operating conditions.
Whether the programme produces technologies capable of competing with established systems remains uncertain. Its immediate significance is in reducing the infrastructure barriers that can prevent new computing approaches from progressing beyond laboratory demonstrations.
