Tensor Processing Unit
Facts
6 properties- Instance of
- processor type
- Has use
- machine learning, multilayer perceptron, convolutional neural network, long short-term memory, TensorFlow, AlphaGo, hardware acceleration[6][7][8][9][10][11][12]
- Different from
- Google Tensor[13]
- Has part(s) of the class
- systolic array[14]
- Image (Wikimedia Commons)
- TPU v4.png ↗
- GitLab.com topic ID
- Tensorflow+Processor+Unit
- Google Knowledge Graph ID
- /g/11byyk3r7m
Has use 7
Subclass of 3
Developer 1
Different from 1
Examples 1
Has part(s) of the class 1
← Different from 1
0 sources
Summary
Tensor Processing Unit is a processor type[1].
high ≥ 0.90 · emerging ≥ 0.65 · low below — calibrated against hand-rated truth (single-source .44 · two-source .62 · production-typical .70), so a band is a measurement, not a promise. High is rare by design: it needs both a supported assertion and an evidence sentence. A confirmed or contradicted chip is a Wikidata cross-check; no chip means Wikidata has no value for that property — the ordinary case here, and it says nothing either way. A confirm-only note marks a property whose check can never disagree (Wikidata's border and capital data is not exhaustive), so silence there is a limit of the check, not of the claim.
evidence “Google’s custom TPU technology is particularly important to its AI operations because the company uses the chips to train and run large AI models at scale.” americanbazaaronline.com · 2026-10-27 · read source
evidence “19, Broadcom shares fell about 5% after Marvell Technology disclosed an expanded custom-chip agreement with Google, whose TPU chips Broadcom has long designed.” finance.yahoo.com · read source
Show 2 low-confidence claims
Published, not hidden — scoring low is not a reason to withhold a sourced claim. A contradicted chip means Wikidata holds a different value; both are shown so you can judge.
evidence “Digital Daily reported that on GKE, customers can set policies to split work between Nvidia GPUs and TPUs, for example falling back to TPU 8i when GPU capacity runs out.” techtimes.com · read source
evidence “However, the chief way clients can access Alphabet's TPUs is through Google Cloud.” fool.com · read source