TPU (Tensor Processing Unit)

TPU (tensor processing unit) is hardware from Google optimized for matrix and tensor computations that dominate deep learning. It is especially useful when training or batch inference needs to scale in Google Cloud environments.

Compare with GPU: both parallelize workloads, but TPUs are more specialized while GPUs are more general-purpose and common in research and third-party stacks. For end users of Gemini, this runs in the background.

---

Key characteristics

• Is specialized hardware optimized for tensor operations in machine learning, especially in Google’s ecosystem.
• Can deliver high efficiency for training and inference when the workload fits the architecture.
• Is most relevant for technical teams deciding runtime environment, performance, and cloud strategy.