Graphcore develops the complete AI compute stack, from custom silicon to datacenter-level infrastructure, centered on its Intelligence Processing Unit (IPU). The IPU is a processor architecturally designed for machine intelligence workloads rather than adapted from GPU or CPU designs. A key engineering milestone was becoming the first company to bring a Wafer-on-Wafer AI processor to market, a packaging approach with implications for interconnect density and yield that the engineering team had to solve at the fabrication level.
The technical domains span semiconductor engineering, processor design, AI model optimization, and the software layers needed to make the hardware usable for machine learning practitioners. This full-stack approach - owning the boundary between silicon and compiler/runtime - is the company's strategy for extracting performance, as opposed to relying on third-party hardware abstractions. The software stack must map the parallelism and memory access patterns of modern AI models onto the IPU's specific architecture, a non-trivial constraint that shapes both product and research directions.
Graphcore is a wholly owned subsidiary of SoftBank Group, backed by long-term investment. The company operates with expanding engineering teams globally, recruiting across semiconductor, systems, and AI software disciplines. Roles involve working on problems that sit at the hardware-software interface: compiler design for novel architectures, systems-level performance optimization, and advancing the capabilities of purpose-built AI silicon.