SoftBank invests $450M in Graphcore’s UK AI chip push

SoftBank has invested more than $450 million in Graphcore, the U.K. company that designs Intelligence Processing Units, or IPUs, for machine learning workloads. The funds are intended to support further IPU development and expand Graphcore’s commercial operations.

Founded in 2016, Graphcore builds processors that target matrix math and model training tasks rather than graphics rendering. The company has focused on hardware and software that aim to run machine learning models efficiently in data centers.

Before the latest investment, Graphcore had raised about $682 million from investors including Sequoia Capital and Microsoft. While the company developed working hardware and software demonstrations, it reported challenges converting technical progress into larger commercial deployments and sustained revenue growth.

SoftBank’s investment follows the company’s broader activity in chip-related assets. SoftBank controls Arm Holdings, the chip architecture firm whose designs are widely licensed across the smartphone industry. Adding Graphcore gives SoftBank an additional asset focused on dedicated AI silicon for data-center compute.

The AI accelerator market is concentrated. Nvidia’s H100 and B200 accelerators command premium pricing and long customer waitlists. Other providers include AMD’s Instinct line, Google’s Tensor Processing Units, Amazon’s Trainium, and specialist vendors such as Cerebras and SambaNova. Intel acquired Habana Labs in 2019. One market hurdle for competitors has been the software ecosystem built around Nvidia’s CUDA platform, which raises switching costs for customers.

Graphcore’s IPUs are designed for machine learning matrix operations and model training rather than cryptocurrency mining or graphics. The new capital is expected to support scaling of manufacturing, development of software tools, and sales efforts for IPU-based systems. Graphcore will need to expand customer adoption, ecosystem support and partnerships to increase deployments of its systems in production environments.

The investment aligns Graphcore’s technical roadmap with SoftBank’s existing holdings in chip architecture and silicon. Companies and cloud providers continue to decide between building custom processors and buying third-party accelerators for machine learning workloads.