Kao to host five AI inference clusters at Harlow by early 2027

Grace Kim

Kao Data will host five high-density GPU (graphics processing unit) clusters at Harlow, with CommonAI's AI Scaling Inference Lab targeting operations across all five by early 2027.

Two dedicated technology suites.

CommonAI selected Kao Data to host the clusters at Harlow. The UK's Advanced Research & Invention Agency (ARIA) funds the lab, while CommonAI delivers it. Its stated purpose is to provide a neutral environment where emerging AI technologies undergo testing and benchmarking against commercial systems.

ARIA launched the lab earlier this year with a £50 million ($66 million) commitment. In June, the UK Government announced more than £20 million ($26 million) in additional funding.

The five-cluster programme covers pipeline parallelism (splitting model stages across linked processors), heterogeneous orchestration (coordinating work across different chip types) and nanosecond optical circuit switching (reconfiguring optical connections at nanosecond speed). Each cluster will test a distinct approach. CommonAI has not shared the hosting locations for the first two clusters.

The hardware mix will include NVIDIA HGX B300, H200 NVL and RTX PRO 6000 Blackwell GPUs, plus AMD Instinct MI350X accelerators. Supermicro, MSI, Fujitsu Primergy and Dell PowerEdge servers will house them.

“Kao Data's Harlow campus is perfectly situated between London and Cambridge, two of the UK's most important AI hubs, making it the ideal home for the Scaling Inference Lab,” said David Stringer, Partnerships Lead, Scaling Inference Lab at CommonAI.

“With advanced levels of security and a distinctive design, the campus provides us with a solid environment to welcome startups, investors and policymakers into the lab, and provides UK innovators with a neutral place to prove their technologies at rack scale - strengthening and supporting the Government's wider AI ambitions.”

What does the Harlow deployment change for UK AI?

For UK AI suppliers, the practical change is access to a government-funded, neutral test environment at rack scale (production-density configurations), with NVIDIA and AMD hardware and optical switching in the test programme. Startups, investors and policymakers can use the lab to evaluate emerging systems against commercial systems.

The first three deployments address inference performance from commodity GPUs, workload distribution across mixed hardware and interconnect delay between linked processors. All five clusters target early 2027 operations.

“Kao Data has a long heritage of hosting some of the UK's most advanced AI and supercomputing workloads, and we are proud to become home to the Scaling Inference Lab,” said Spencer Lamb, Kao Data CEO.

“The work taking place at our Harlow campus will help shape the next generation of UK AI, ensuring the country remains a leader in AI development while meeting the growing demand for sovereign, homegrown AI infrastructure and the resources needed to support the country's world-leading startup community.”

Deployment schedule and test targets

Cluster 1's build-out began earlier this year with hot aisle containment (a cooling configuration that captures hot exhaust), and the cluster launched in June. It will run until April 2027 while testing whether pipeline parallelism can unlock meaningful inference performance from commodity GPU hardware.

Cluster 2 is scheduled to deploy in August 2026. It will test whether intelligent workload distribution (assigning jobs to suitable processors) can optimize performance across mixed hardware. Cluster 3 will examine whether lower interconnect latency (delay between linked processors) from optical switching produces meaningful gains in end-to-end AI inference throughput.

Technology partners are Quettaflop AI, Callosum and Oriole Networks.

Harlow capacity and expansion

Kao Data launched Harlow in 2018 on a plan for four 10MW buildings. Kao Data operates KLON-01 and KLON-02, each offering 10MW. Work on the second data center began in March 2022 and finished in late 2023.

In March 2025, Kao Data announced KLON-03, a 17.6MW facility. In September 2025, Kao Data revealed plans for two further data centers, taking the campus total to five buildings. NVIDIA's Cambridge-1 supercomputer sits within the original Harlow facility, and European neocloud (specialist cloud provider) Nebius is taking capacity at the site.

Kao Data's wider footprint

Kao Data launched in 2015 and has three owners: Infratil, Legal & General and Goldacre. Its portfolio spans more than 237MW of IT load (power available to computing equipment) that is operational, under development or planned.

Besides Harlow, Kao Data operates a Slough data center and the former Barclays data center in Northolt. It is developing a Manchester campus and recently bought an industrial site in London's Park Royal for data center development.

Kao Data takes its name from Sir Charles Kao, who won the Nobel Prize in Physics for creating optical fiber cables while working at the Standard Telephone Laboratories (STL) research center in Harlow. Those laboratories stood on the land now occupied by Kao Data's data centers.