AI data center builder Nscale secures $155 million investment

Nscale Ltd., a London startup that builds data centers optimized for artificial intelligence workloads, has raised $155 million to increase its infrastructure footprint.

The Series A round has taken place announcement This morning. Sandton Capital Partners led the investment with participation from Kestrel 0x1, Blue Sky Capital Managers and Florence Capital.

This funding milestone comes weeks after an AI cluster built by Nscale entered the Top500 ranking of the world’s most powerful supercomputers. The Svartisen cluster caught 156th place with 12.38 petaflops of peak performance and 66,528 cores.

Nscale assembled the system from servers each comprising six Advanced Micro Devices Inc. chips: two central processing units and four MI250X machine learning accelerators. The MI250X includes two graphics cards based on a six-nanometer process, as well as 128 GB of onboard RAM for storing data from AI models.

The servers are linked together by an Ethernet network built by Nscale using Broadcom Inc. silicon. The network implements a technology called RoCE that can move data between two machines in a way that bypasses their respective processors, speeding up the traffic flow. RoCE also automates tasks such as detecting overloaded network links and redirecting packets to other connections.

At the software level, Nscale’s hardware is powered by a custom infrastructure orchestration platform. It integrates Kubernetes with Slurm, a popular open source framework for managing data center infrastructure.

Both Kubernetes and Slurm automate the task of determining which workload should run on which server in a cluster. However, they differ on several points. Kubernetes has a self-healing mechanism, which allows it to automatically recover from certain types of malfunctions. Slurm, in turn, supports a networking technology called MPI that can move data between different components of an AI workload with a high degree of efficiency.

Nscale built the Svartisen cluster in Glomfjord, a Norwegian village located inside the Arctic Circle. The data center (pictured) that hosts the system is powered by a nearby hydroelectric dam and sits directly on a fiber optic cable that connects it to the internet service providers’ infrastructure. The cable features dual redundancy, meaning it can continue to function even if several important components fail.

The company makes its infrastructure available to its customers in several ways. It provides AI training clusters along with an inference service that automatically adds or removes hardware resources based on workload demand. There are also bare metal infrastructure options, which allow users to more extensively customize the software stack that powers their deployments.

Customers can download AI models from an algorithm library provided by Nscale or contribute their own. According to the company, there is a preconfigured compiler toolkit to transform user workloads into a form that can run efficiently on its servers.

Currently, Nscale’s construction pipeline includes data centers with an overall power consumption of 300 megawatts. This is 10 times the amount of electricity used by the company’s Glomfjord facility. With the Series A funding round announced today, Nscale will increase its pipeline by 1,000 megawatts.

“The biggest risk to the market’s ability to scale lies in the large contiguous slices of electricity needed to power these large GPU superclusters,” said Joshua Payne, CEO of Nscale. “Nscale has a 1.3 GW site pipeline in its portfolio, enabling us to design the end-to-end data center, supercluster and cloud environment for our customers. »

The company will build the new data centers in North America and Europe. According to the company, data center capacity worth 120 megawatts is expected to be built next year. The new infrastructure will help Nscale power a planned public cloud offering focused on training and inference workloads, which is expected to launch in the first quarter of 2025.

Photo: Nescale