Graphcore was founded in Bristol in 2016 by Nigel Toon and Simon Knowles, both veterans of the UK semiconductor industry. The founding premise was straightforward: the GPU, designed originally for graphics rendering, was a poor architectural fit for machine learning workloads. Toon and Knowles set out to build a processor designed from first principles around the computational patterns of AI, producing what they called the Intelligence Processing Unit (IPU).
The company raised substantial venture backing through the late 2010s and early 2020s, attracting investment from firms including Sequoia and BMW i Ventures, and at its peak was valued at roughly $2.8 billion, making it one of the most prominent deep-tech unicorns to emerge from the UK. Its IPU architecture differs from conventional GPU design by distributing memory across the processor itself rather than relying on high-bandwidth external memory, a trade-off that offers advantages on certain sparse, irregular AI workloads.
Graphcore operates in a market that Nvidia has come to dominate with considerable force. That context matters. The company's trajectory illustrates a persistent tension in the AI hardware space: architectural innovation is genuinely possible, but the software ecosystem, tooling, and developer familiarity that surrounds an incumbent can outweigh raw silicon performance in purchasing decisions. For operators and founders watching the AI infrastructure layer, Graphcore is a useful case study in what it takes to challenge a platform monopoly, and how difficult the gap between technical credibility and commercial scale can be to close.