TPU
In plain wordsGoogle's own family of chips built specifically for AI, used to train and run its Gemini models and rented to other companies through its cloud.
Google's family of application-specificASIC, abbr.: A chip whose circuits are laid out to do one kind of work extremely well instead of running any program. AI companies design their own to run their models more cheaply than with off-the-shelf chips. AI acceleratorsAI accelerator, n.: Any chip built or tuned to speed up AI — graphics chips, Google's TPUs, phone 'neural engines' and many startup designs. Such chips do the heavy math of training and running models., built around large systolicsystolic array, n.: A grid of tiny calculators that pass numbers to their neighbors in rhythm, like a heartbeat, so each value is reused many times. AI chips use it to multiply large tables of numbers efficiently. matrix units. The first TPU, in Google's data centers from 2015, was an inference chip whose 256 × 256 array of 65,536 8-bit multiply–accumulate units delivered 92 trillion operations per second. TPU v2 (2017) added training and introduced the bfloat16bfloat16, n.: A 16-bit number format invented at Google for AI. It keeps the huge range of the standard 32-bit format but drops precision, which neural networks can usually tolerate. format; later generations pair high-bandwidth memoryhigh-bandwidth memory, n.: Very fast memory made by stacking memory chips on top of each other and placing them right next to the processor, so AI chips can be fed data quickly enough. with a dedicated inter-chip interconnect and, from TPU v4 (2021), optical circuit switches that link thousands of chips into reconfigurable 'pods'. The seventh generation, Ironwood (2025), offers about 4.6 FP8 petaFLOPS and 192 GB of HBM per chip in pods of up to 9,216 chips, and two eighth-generation chips, TPU 8t and TPU 8i, were announced in April 2026. TPUs are programmed through XLAXLA, abbr.: A compiler from Google that turns AI programs into fast code for chips such as TPUs and GPUs, mainly by merging many small steps into fewer, bigger ones. from JAXJAX, prop. n.: A free Google library that lets researchers write mathematics in familiar Python and automatically get derivatives and very fast code for graphics chips and Google's AI chips., TensorFlowTensorFlow, prop. n.: Google's free software toolkit for building, training and running machine-learning models; it led the field after 2015 and still runs many models inside phones, browsers and company systems. or PyTorchPyTorch, prop. n.: The most widely used free software toolkit for building and training neural networks, especially in research; programmers write ordinary Python and it handles the calculus and the graphics chips.. Besides training Google's Gemini models, they are rented through Google Cloud; in October 2025 Anthropic agreed to use up to one million of them.
Origin
Designed from 2013, when Google projected that neural-network speech recognition could double its data-center needs; deployed in 2015 and announced at Google I/O in May 2016. Norman Jouppi and 74 co-authors described the first chip at ISCA 2017. The name nods to TensorFlowTensorFlow, prop. n.: Google's free software toolkit for building, training and running machine-learning models; it led the field after 2015 and still runs many models inside phones, browsers and company systems..
Usage note
'TPU' names only Google's line, though it is sometimes misused for AI chips generally. The Edge TPU is a separate small inference chip for devices.
Related
- AI acceleratorAI accelerator, n.: Any chip built or tuned to speed up AI — graphics chips, Google's TPUs, phone 'neural engines' and many startup designs. Such chips do the heavy math of training and running models.
- systolic arraysystolic array, n.: A grid of tiny calculators that pass numbers to their neighbors in rhythm, like a heartbeat, so each value is reused many times. AI chips use it to multiply large tables of numbers efficiently.
- ASICASIC, abbr.: A chip whose circuits are laid out to do one kind of work extremely well instead of running any program. AI companies design their own to run their models more cheaply than with off-the-shelf chips.
- XLAXLA, abbr.: A compiler from Google that turns AI programs into fast code for chips such as TPUs and GPUs, mainly by merging many small steps into fewer, bigger ones.
- bfloat16bfloat16, n.: A 16-bit number format invented at Google for AI. It keeps the huge range of the standard 32-bit format but drops precision, which neural networks can usually tolerate.
- GPUGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time.
- AI supercomputerAI supercomputer, n.: A huge network of AI chips — often tens or hundreds of thousands — wired together in a data center to train the largest AI models.
- GPUGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time.
Sources
- Jouppi, N. P. et al. (2017). In-Datacenter Performance Analysis of a Tensor Processing Unit. Proceedings of ISCA 2017. arxiv.org/abs/1704.04760
- Jouppi, N. P. et al. (2023). TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings. Proceedings of ISCA 2023.
Mentioned in
AI acceleratorAI accelerator, n.: Any chip built or tuned to speed up AI — graphics chips, Google's TPUs, phone 'neural engines' and many startup designs. Such chips do the heavy math of training and running models., computecompute, n.: The raw computing power used to train and run AI, usually measured by how many specialized chips are used and for how long. With data and algorithms, it is a main driver of AI progress., Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold., GPUGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time., NvidiaNvidia, prop. n.: An American chip company whose graphics processors became the main hardware for training and running AI, making it one of the most valuable companies in the world., AI stackAI stack, n.: The layers of technology an AI product depends on, from chips and data centers at the bottom, through the models, up to the apps people use, each built on the layer below., AlphaZeroAlphaZero, prop. n.: A 2017 DeepMind program that taught itself chess, shogi and Go from nothing but the rules by playing against itself, and within hours of training beat the strongest existing programs., architecturearchitecture, n.: The overall design of an AI model, meaning what kinds of parts it has and how they are connected, as opposed to the specific numbers it learns during training., ASICASIC, abbr.: A chip whose circuits are laid out to do one kind of work extremely well instead of running any program. AI companies design their own to run their models more cheaply than with off-the-shelf chips., Attention Is All You NeedAttention Is All You Need, prop. n.: The 2017 Google research paper that introduced the transformer, the neural-network design behind ChatGPT, Claude, Gemini and nearly every other modern AI language system., bfloat16bfloat16, n.: A 16-bit number format invented at Google for AI. It keeps the huge range of the standard 32-bit format but drops precision, which neural networks can usually tolerate., hyperscalerhyperscaler, n.: One of the few giant technology companies — such as Amazon, Microsoft, Google and Meta — that run vast networks of data centers and spend most of the money that builds AI computing., INT8INT8, abbr.: Storing numbers as whole numbers from −128 to 127 using 8 bits. AI models converted to this format run faster and need a quarter of the memory of 32-bit versions., Jeff DeanJeff Dean, prop. n.: An American computer scientist who built much of Google's computing infrastructure, co-founded the Google Brain deep-learning team and was Google's chief scientist before leaving in 2026 to co-found a startup., matrixmatrix, n.: A rectangular grid of numbers arranged in rows and columns. Most of what a neural network has learned is stored in large grids like this., matrix multiplicationmatrix multiplication, n.: The rule for combining two grids of numbers into a new grid by repeatedly multiplying rows by columns. It is the single calculation AI chips spend most of their time doing., tensor coretensor core, n.: A small, specialized part of an AI chip that does one job — multiplying small blocks of numbers together — far faster than ordinary circuits can., TensorFlowTensorFlow, prop. n.: Google's free software toolkit for building, training and running machine-learning models; it led the field after 2015 and still runs many models inside phones, browsers and company systems., training runtraining run, n.: One complete attempt at training a model with a fixed recipe, from start to finish. For the largest AI models a single run can occupy tens of thousands of chips for months., TSMCTSMC, prop. n.: The Taiwanese company that manufactures most of the world's advanced computer chips, including nearly all of the specialized chips used to train and run AI., vLLMvLLM, prop. n.: A free, open-source program for running large language models on servers efficiently, so one machine can answer many more users at once. Many companies use it to host AI models., JAXJAX, prop. n.: A free Google library that lets researchers write mathematics in familiar Python and automatically get derivatives and very fast code for graphics chips and Google's AI chips., orbital data centerorbital data center, n.: A proposed computer center circling the Earth, powered by almost constant sunlight, that would run AI without using land, water or power grids on the ground., PaLMPaLM, prop. n.: A very large Google language model from 2022 that showed sudden jumps in reasoning, and even explained jokes, as it grew. Its successor powered Google's Bard chatbot before the Gemini models replaced it.