Graphcore

Nigel Toon and Simon Knowles founded Graphcore in Bristol in 2016 with a technically precise thesis: the dominant chip for machine learning — the GPU, originally designed for graphics rendering — is not the optimal architecture for modern AI workloads. Neural network models have computational characteristics that differ fundamentally from the tasks GPUs were built for, and a new processor architecture, designed from the ground up for artificial intelligence, could outperform them significantly. Graphcore developed the Intelligence Processing Unit, or IPU — a processor with a fundamentally different architecture from conventional GPUs. Where a GPU is optimized for dense, regular operations on large matrices, the IPU is designed to handle the irregular, sparse, and massively parallel computation that characterizes the training and inference of modern neural networks, particularly transformer models and large language models. The addressable market is substantial and growing fast. Demand for AI compute has exceeded projections year after year, driven by the explosion of large language models and by the investment of major technology companies, research institutions, and governments in AI infrastructure. Nvidia currently dominates this market, but the concentration of that dominance has pushed many organizations to actively seek capable alternatives. Graphcore's business model combines hardware sales — IPU boards and rack systems — with a proprietary software platform called Poplar, which allows researchers and engineers to compile and optimize their models for the IPU architecture. The company targets the premium end of AI compute: research centers, universities, and companies working at the cutting edge of machine learning, where the difference between architectures is both measurable and commercially significant. Graphcore has raised hundreds of millions of dollars from top-tier investors and built a customer base that includes leading research institutions in Europe and beyond. It represents one of the most significant bets in the British deep tech ecosystem on the possibility that the future of AI compute will not belong exclusively to Nvidia.