TPU (Tensor Processing Unit) is a type of computer chip designed by Google specifically to accelerate machine learning workloads. It is purpose-built for the maths that powers neural networks.
What it means in plain English
Where GPUs were adapted from graphics to AI, TPUs were designed from the ground up for it. They are optimised for the specific operations neural networks rely on, which can make them very efficient for training and running large models. TPUs are used heavily within Google’s own AI systems and are available to others through its cloud platform.
They are one example of the broader trend toward specialised AI hardware, as demand for computing power grows.
A simple example
When Google trains or runs its large AI models, it often uses TPUs. A developer building an AI application on Google’s cloud can also rent TPU power to train their own models faster.
Why it matters
TPUs illustrate how important specialised hardware has become to AI. As models grow, purpose-built chips like TPUs help make training and running them faster and more efficient — a key factor in who can build cutting-edge AI.
Related terms
- GPU — the more general AI chip TPUs are an alternative to.
- Deep Learning — the workload TPUs accelerate.
- Inference — one task TPUs can speed up.
Frequently asked questions
What is a TPU?
A Tensor Processing Unit is a chip designed by Google specifically to accelerate machine-learning workloads, especially neural-network training and inference.
How does a TPU differ from a GPU?
Both accelerate AI, but TPUs are purpose-built for tensor operations and are offered mainly through Google Cloud, while GPUs are more general-purpose and widely available.
