Nscale Ltd., a London-based startup that creates data centers for buider AI projects, has raised $155 million in its Series A funding round. The funding was led by Sandton Capital Partners, with support from Kestrel 0x1, Blue Sky Capital Managers, and Florence Capital. The announcement comes right after Nscale’s Svartisen Cluster was ranked 156th on the Top500 list of the world’s most powerful supercomputers, with a performance of 12.38 petaflops and 66,528 cores.   

Technology and Infrastructure

• Nscale built the Svartisen Cluster using servers with six AMD chips: two central processing units (CPUs) and four MI250X machine learning accelerators.
• The MI250X chips have two graphics cards built with a six-nanometer process and 128GB of memory to store AI model data.
• The servers are connected through an Ethernet network that Nscale created using Broadcom Inc. technology. The network uses RoCE (RDMA over Converged Ethernet), which makes data transfer faster by bypassing the CPUs of the connected machines. RoCE also automatically manages network traffic to prevent delays caused by overloaded connections.

Software and Management

• Nscale’s hardware operates on a custom platform that controls the entire data center system.
• The platform combines Kubernetes and Slurm — two tools that help manage tasks and assign servers. Kubernetes has a self-healing feature that automatically fixes issues, while Slurm uses MPI (Message Passing Interface) to move data quickly between different parts of an AI task.

Location and Power Supply

• Nscale built its Svartisen Cluster in Glomfjord, a small village in Norway, located inside the Arctic Circle.
• The data center is powered by a nearby hydroelectric dam and connected to the internet via a fiber-optic cable with double backups. This means the system can keep running even if some parts stop working.

   Read this link also :   https://sifted.eu/articles/tech-events-2025

Services for Customers

• AI training clusters.
• Inference services that automatically change hardware resources depending on the workload.
• Bare-metal infrastructure, which allows users to customize the software on their systems.
• Customers can either download AI models from Nscale’s library or upload their own models. Nscale also provides a pre-made compiler toolkit to help optimize user tasks, making sure they run smoothly on its system.
• For those looking to build custom AI solutions, Nscale offers flexible, high-performance infrastructure. This AI platform helps users enhance and grow their custom models.

Future Expansion and Plans

• Nscale’s current infrastructure includes data centers that use 300 megawatts of power, which is 10 times more than the Glomfjord facility.
• With the Series A funding, Nscale plans to increase its capacity by 1,000 megawatts.
• Nscale CEO Joshua Payne explains that a major challenge in growing the market is the large, continuous power needed to run GPU superclusters.
• Nscale has a 1.3GW pipeline of sites, which allows the company to build everything from scratch—data centers, superclusters, and cloud environments—to fit customer needs.
• New data centers will be built in North America and Europe. Nscale also plans to add 120 megawatts next year.
• The new infrastructure will support the launch of Nscale's public cloud service for AI training and inference, which is set to go live in the first quarter of 2025.