COMPEL Glossary / tpu-tensor-processing-unit
TPU (Tensor Processing Unit)
A TPU is a custom-designed processor created by Google specifically for neural network workloads, available through Google Cloud Platform.
What this means in practice
TPUs are optimized for the tensor (multi-dimensional array) operations that are fundamental to training and running transformer-based AI models. They offer competitive performance for specific workload types, particularly large-scale model training and inference. For transformation leaders evaluating cloud infrastructure options, TPUs represent one point on the growing spectrum of AI-specific hardware -- alongside GPUs from NVIDIA and AMD, AWS's custom Trainium and Inferentia chips, and processors from startups like Cerebras and Graphcore. Infrastructure decisions are among the most consequential in AI transformation because they involve multi-year commitments with compounding cost and capability implications.
Related Terms
Other glossary terms mentioned in this entry's definition and context.