Google’s Tensor Processing Unit (TPU) is a specialised AI accelerator originally built for large-scale cloud computing in the company’s data centres, while GPUs are more general-purpose processors used for graphics, AI training and other demanding tasks. The distinction matters because Google also calls the AI hardware in its Pixel 11 phones a TPU, although in that setting it effectively serves the role normally associated with a neural processing unit (NPU).

Data-centre TPUs are designed for massive AI and machine-learning workloads, such as training large language models, using a systolic-array layout that passes calculations directly between units rather than repeatedly accessing memory. This can reduce bottlenecks, latency and energy use compared with GPUs for enormous workloads; meanwhile, NPUs and phone-based TPUs handle local tasks such as image processing and generative AI. Google says the Pixel 11’s Tensor G6 offers 50% more TPU computing power and up to 3.5 times faster AI processing while using up to 3.5 times less energy.