product
Google TPU (Tensor Processing Unit)
Also known as TPU, Tensor Processing Unit, Cloud TPU, Ironwood, TPU 8t, TPU 8i
The Tensor Processing Unit (TPU) is Google's custom chip for machine learning, an ASIC built for the matrix maths in AI models.[1] Google has used TPUs in its data centers since 2015.[2] In April 2026 it split the eighth generation into two chips, TPU 8t for training and TPU 8i for inference.[3] Google now sells TPU systems to some customers for their own data centers, not only renting them through Google Cloud.[4]
Key facts
Google’s Tensor Processing Units are custom application-specific integrated circuits (ASICs) built to accelerate machine learning, especially the large matrix operations at the core of neural networks.[1] They power Google’s own models, including Gemini, and are rented and, since 2026, sold to outside customers.[5][4] For how TPUs compare with GPUs and NPUs, see the AI chips explainer.
Origins: an inference chip
Google deployed its first TPU in its data centers in 2015 to speed up the inference phase of neural networks, meaning running models that were already trained.[2] The chip was built around a 65,536-unit 8-bit matrix multiply array with a peak of 92 trillion operations per second and 28 MiB of on-chip memory.[6] In a 2017 paper, Google engineers reported it was on average 15 to 30 times faster than a contemporary NVIDIA K80 GPU or Intel Haswell CPU on Google’s production inference workloads, with 30 to 80 times better performance per watt.[7] The paper argued that big gains in cost, energy and performance would now have to come from domain-specific hardware.[8]
Google counts TPU v2, from 2018, as its first Cloud TPU available to outside customers.[9]
Ironwood, the seventh generation
In April 2025 Google introduced Ironwood, its seventh-generation TPU, which it called the first TPU designed specifically for inference.[10] Google said Ironwood scales to pods of 9,216 liquid-cooled chips with 42.5 exaflops of compute, carries 192 GB of high-bandwidth memory (HBM) per chip at 7.37 TB/s, and doubles performance per watt over the sixth-generation Trillium.[11] Google said Gemini 2.5 and AlphaFold ran on TPUs.[5]
The eighth generation: TPU 8t and TPU 8i
On April 22, 2026, at Google Cloud Next, Google introduced its eighth-generation TPU as two chips: TPU 8t for training and TPU 8i for inference.[3] Tom’s Hardware noted it was the first time the program had shipped two distinct designs in one generation.[12] Google said it anticipated rising inference demand and the needs of AI agents, so TPU 8t gets more compute and scale-up bandwidth while TPU 8i gets more memory bandwidth, though both can run either kind of work.[13] Both were designed with Google DeepMind.[14]
- TPU 8t. A superpod scales to 9,600 chips with two petabytes of shared HBM and 121 exaflops, nearly three times the compute per pod of Ironwood.[15] Google says its Virgo Network and Pathways software let it scale near-linearly to a million chips in one logical cluster.[16]
- TPU 8i. Each chip pairs 288 GB of HBM with 384 MB of on-chip SRAM, three times the previous generation.[17] Interconnect bandwidth doubles to 19.2 Tb/s, a new Boardfly topology cuts the maximum network diameter by more than half, and a Collectives Acceleration Engine cuts on-chip latency by up to five times.[18]
- Claimed gains. Google claims up to 2.7 times better training performance per dollar for TPU 8t, up to 80% better inference performance per dollar for TPU 8i, and up to twice Ironwood’s performance per watt for both.[19] These are Google’s own claims.[19]
- Host and software. Both are the first TPUs to run on Google’s Arm-based Axion CPUs, and both support JAX, PyTorch, SGLang and vLLM.[20]
Google said both chips would be generally available later in 2026.[21] Per chip, Tom’s Hardware put TPU 8t at 12.6 FP4 petaflops and TPU 8i at 10.1, roughly a third of NVIDIA’s Vera Rubin, but argued Google’s advantage lies in pod-level scale.[22][23] At the same event Google also announced NVIDIA Vera Rubin instances.[24]
Design partners and manufacturing
Google designs TPUs with chip partners. In April 2026 Broadcom disclosed a long-term agreement to develop and supply custom TPUs for Google’s future generations, plus networking and other rack components, through up to 2031.[25] Tom’s Hardware reported that MediaTek joined as a second design partner in December 2025, ending Broadcom’s exclusive role since 2015, and that MediaTek is understood to handle TPU 8i while Broadcom handles TPU 8t.[26] The same outlet reported both chips are made on TSMC‘s N3 process family with HBM3E memory.[27]
External customers
TPUs were historically available only for rent through Google Cloud.[28] That changed in 2026. Alphabet says it has signed a limited number of agreements to supply TPU systems to customers’ on-premises infrastructure, began recognizing revenue from them in the second quarter of 2026, and expects most of it in 2027.[4] Google Cloud revenue grew 103% to $30.7 billion in the third quarter of 2026, with TPU system sales a meaningful contributor.[29]
Anthropic is one of the named customers. In October 2025 it said it would expand to up to one million TPUs.[30] Broadcom’s April 2026 filing said Anthropic will access about 3.5 gigawatts of next-generation TPU capacity through Broadcom from 2027.[31] Google also named Citadel Securities as a TPU user.[32] Meta reportedly agreed in February 2026 to rent Google Cloud TPUs, according to The Information as relayed by SiliconANGLE.[33] Whether these deals dent NVIDIA’s position is covered in the custom chips debate.
Read next
NVIDIA
NVIDIA designs GPUs, CPUs and networking for AI data centers. Its history, products, finances and recent developments as of November 2026.
Questions readers ask
What is a TPU?
A Tensor Processing Unit is Google's custom application-specific integrated circuit for machine learning. It is built for large matrix operations and has on-chip high-bandwidth memory.[1]
What are TPU 8t and TPU 8i?
They are Google's eighth-generation TPUs, announced on April 22, 2026. TPU 8t is built for training, with more compute and scale-up bandwidth. TPU 8i is built for inference, with more memory bandwidth. Google said both would be generally available later in 2026.[3][13][21]
Who makes Google's TPUs?
Google designs TPUs with partners. Broadcom has a long-term agreement to develop and supply future TPU generations. Tom's Hardware reported that MediaTek also joined as a design partner and that both eighth-generation chips are made on TSMC's N3 process.[25][26][27]
Can companies other than Google use TPUs?
Yes. Cloud customers can rent them, and since the second quarter of 2026 Google has recognized revenue from a limited number of agreements to supply TPU systems for customers' own data centers. Anthropic has committed to multiple gigawatts of TPU capacity.[4][31]
Sources
· 34 claims from 10 publishers
Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.
- [1]
Google describes TPUs as its custom-developed application-specific integrated circuits (ASICs) for accelerating machine learning, built for the large matrix operations common in machine learning and equipped with on-chip high-bandwidth memory (HBM). confirmedas of 2026-10-26
- Introduction to Cloud TPU · Google Cloud documentation · Primary source (retrieved 2026-11-07)
- Introduction to Cloud TPU · Google Cloud documentation · Primary source (retrieved 2026-11-07)
- Introduction to Cloud TPU · Google Cloud documentation · Primary source (retrieved 2026-11-07)
- [2]
Google's first TPU was a custom ASIC deployed in its data centers from 2015 to accelerate the inference phase of neural networks. confirmedas of 2017-04-16
- In-Datacenter Performance Analysis of a Tensor Processing Unit · arXiv (Jouppi et al., Google) · Primary source · 2017-04-16 · Abstract (retrieved 2026-11-07)
- [3]
On April 22, 2026, at Google Cloud Next, Google introduced its eighth-generation TPU as two separate chips, TPU 8t for training and TPU 8i for inference. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [4]
Alphabet says Google Cloud generates product revenue primarily from selling TPU systems, and that it has signed a limited number of agreements to supply TPU systems for customers' on-premises infrastructure, recognizing revenue from them since the second quarter of 2026 with most to be recognized in 2027. confirmedas of 2026-09-30
- Alphabet Inc. Form 10-Q for the quarter ended September 30, 2026 · Alphabet (SEC filing) · Primary source · 2026-10-29 (retrieved 2026-11-07)
- Alphabet Inc. Form 10-Q for the quarter ended September 30, 2026 · Alphabet (SEC filing) · Primary source · 2026-10-29 (retrieved 2026-11-07)
- Alphabet Inc. Form 10-Q for the quarter ended September 30, 2026 · Alphabet (SEC filing) · Primary source · 2026-10-29 (retrieved 2026-11-07)
- [5]
Google said in April 2025 that Gemini 2.5 and AlphaFold run on TPUs. confirmedas of 2025-04-09
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- [6]
The first TPU was built around a 65,536-unit 8-bit multiply-accumulate matrix unit with a peak of 92 trillion operations per second and 28 MiB of software-managed on-chip memory. confirmedas of 2017-04-16
- In-Datacenter Performance Analysis of a Tensor Processing Unit · arXiv (Jouppi et al., Google) · Primary source · 2017-04-16 · Abstract (retrieved 2026-11-07)
- [7]
Google's 2017 paper reported that on its production inference workloads the first TPU was on average about 15 to 30 times faster than a contemporary Nvidia K80 GPU or Intel Haswell CPU, with 30 to 80 times higher performance per watt. confirmedas of 2017-04-16
- In-Datacenter Performance Analysis of a Tensor Processing Unit · arXiv (Jouppi et al., Google) · Primary source · 2017-04-16 · Abstract (retrieved 2026-11-07)
- In-Datacenter Performance Analysis of a Tensor Processing Unit · arXiv (Jouppi et al., Google) · Primary source · 2017-04-16 · Abstract (retrieved 2026-11-07)
- [8]
The 2017 TPU paper argued that many architects believe major improvements in cost, energy and performance must now come from domain-specific hardware. confirmedas of 2017-04-16
- In-Datacenter Performance Analysis of a Tensor Processing Unit · arXiv (Jouppi et al., Google) · Primary source · 2017-04-16 · Abstract (retrieved 2026-11-07)
- [9]
Google refers to its first Cloud TPU, TPU v2, as dating from 2018 and says Ironwood is nearly 30 times more power-efficient. confirmedas of 2025-04-09
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- [10]
In April 2025 Google introduced Ironwood, its seventh-generation TPU, which it called the first TPU designed specifically for inference. confirmedas of 2025-04-09
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- [11]
Google said Ironwood scales to pods of 9,216 liquid-cooled chips delivering 42.5 exaflops, with 192 GB of HBM per chip at 7.37 TB/s, and twice the performance per watt of Trillium, its sixth-generation TPU. confirmedas of 2025-04-09
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- Ironwood: The first Google TPU for the age of inference · Google (The Keyword) · Primary source · 2025-04-09 (retrieved 2026-11-07)
- [12]
Tom's Hardware reported that the eighth generation is the first time in the TPU program's history that Google has shipped two distinct chip designs. confirmedas of 2026-04-27
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [13]
Google says it split the generation because it anticipated rising inference demand and AI agents; TPU 8t has more compute throughput and scale-up bandwidth for training, while TPU 8i has more memory bandwidth for latency-sensitive inference, though both can run either kind of workload. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [14]
Google says TPU 8t and TPU 8i were designed in partnership with Google DeepMind. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [15]
Google says a TPU 8t superpod scales to 9,600 chips with two petabytes of shared HBM and 121 exaflops of compute, nearly three times the compute per pod of the previous generation. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [16]
Google says its Virgo Network, JAX and Pathways software let TPU 8t scale near-linearly to up to a million chips in a single logical cluster, and that TPU 8t targets over 97% goodput. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [17]
Google says TPU 8i pairs 288 GB of HBM with 384 MB of on-chip SRAM, three times the previous generation, to keep a model's active working set on the chip. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [18]
Google says TPU 8i doubles interconnect bandwidth to 19.2 Tb/s, uses a new Boardfly topology that cuts maximum network diameter by more than 50%, and adds a Collectives Acceleration Engine that reduces on-chip latency by up to five times. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [19]
Google claims TPU 8t delivers up to 2.7 times the training performance per dollar of Ironwood and TPU 8i up to 80% better inference performance per dollar, and that both offer up to twice Ironwood's performance per watt. confirmedas of 2026-04-22
- TPU 8t and TPU 8i technical deep dive · Google Cloud Blog · Primary source · 2026-04-22 (retrieved 2026-11-07)
- TPU 8t and TPU 8i technical deep dive · Google Cloud Blog · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [20]
Google says both eighth-generation chips are the first TPUs to run on its own Axion Arm-based CPU hosts, and support JAX, MaxText, PyTorch, SGLang and vLLM with bare-metal access. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [21]
Google said in April 2026 that both eighth-generation TPUs would be generally available later in 2026. confirmedas of 2026-04-22· forecast
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [22]
Tom's Hardware, citing Google's technical deep dive, reported TPU 8t at 12.6 FP4 petaflops with 216 GB of HBM3e and TPU 8i at 10.1 FP4 petaflops with 288 GB of HBM3e, roughly a third of the per-chip compute of Nvidia's Vera Rubin or AMD's MI455X. reportedas of 2026-04-27
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- [23]
Tom's Hardware argued that individual Nvidia GPUs are faster, but Google holds an advantage in pod-level throughput, since Nvidia's current GPUs top out at 576 accelerators in a single NVLink deployment. reportedas of 2026-04-27· interpretation
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- [24]
Google also announced Nvidia Vera Rubin NVL72 instances on the same Virgo Network fabric at Cloud Next 2026. reportedas of 2026-04-27
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- [25]
In April 2026 Broadcom disclosed a long-term agreement to develop and supply custom TPUs for Google's future TPU generations, plus a supply agreement for networking and other components for Google's next-generation AI racks, through up to 2031. confirmedas of 2026-04-06
- Broadcom Inc. Form 8-K (Google TPU long-term agreement and Anthropic capacity), April 6, 2026 · Broadcom (SEC filing) · Primary source · 2026-04-06 (retrieved 2026-11-07)
- Broadcom Inc. Form 8-K (Google TPU long-term agreement and Anthropic capacity), April 6, 2026 · Broadcom (SEC filing) · Primary source · 2026-04-06 (retrieved 2026-11-07)
- [26]
Tom's Hardware reported that MediaTek joined Broadcom as a TPU design partner in December 2025, ending Broadcom's exclusive role since 2015, and that MediaTek is understood to design TPU 8i while Broadcom designs TPU 8t. reportedas of 2026-04-27
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- [27]
Tom's Hardware reported that both eighth-generation TPUs are made on TSMC's N3 process family with HBM3E memory. reportedas of 2026-04-27
- Inside Google's TPU V8 strategy, delivering two chips for two crucial tasks at incredible scale · Tom's Hardware · Secondary source · 2026-04-27 (retrieved 2026-11-07)
- [28]
The Register and SiliconANGLE noted that Google's TPUs had historically been available only for rent through Google Cloud. confirmedas of 2026-02-26
- Nvidia scoffs at threat from Google TPUs · The Register · Secondary source · 2025-11-25 (retrieved 2026-11-07)
- Google and Meta reportedly strike new, multibillion-dollar AI chip deal · SiliconANGLE · Secondary source · 2026-02-26 (retrieved 2026-11-07)
- [29]
Alphabet reported that Google Cloud revenue grew 103% to $30.7 billion in the third quarter of 2026, with TPU systems sales a meaningful contributor, and Sundar Pichai said Google was seeing strong demand for TPU systems across a wide range of customers. confirmedas of 2026-10-28
- Alphabet announces third quarter 2026 results (Form 8-K, Exhibit 99.1) · Alphabet (SEC filing) · Primary source · 2026-10-28 (retrieved 2026-11-07)
- Alphabet announces third quarter 2026 results (Form 8-K, Exhibit 99.1) · Alphabet (SEC filing) · Primary source · 2026-10-28 (retrieved 2026-11-07)
- Alphabet announces third quarter 2026 results (Form 8-K, Exhibit 99.1) · Alphabet (SEC filing) · Primary source · 2026-10-28 (retrieved 2026-11-07)
- [30]
In October 2025 Anthropic said it would expand to up to one million Google TPUs and that it uses three chip platforms: Google TPUs, Amazon Trainium and NVIDIA GPUs. confirmedas of 2025-10-23
- Expanding our use of Google Cloud TPUs and Services · Anthropic · Primary source · 2025-10-23 (retrieved 2026-10-10)
- Expanding our use of Google Cloud TPUs and Services · Anthropic · Primary source · 2025-10-23 (retrieved 2026-10-10)
- [31]
Broadcom said Anthropic will access about 3.5 gigawatts of next-generation TPU-based compute through Broadcom beginning in 2027, with consumption dependent on Anthropic's continued commercial success. confirmedas of 2026-04-06
- Broadcom Inc. Form 8-K (Google TPU long-term agreement and Anthropic capacity), April 6, 2026 · Broadcom (SEC filing) · Primary source · 2026-04-06 (retrieved 2026-11-07)
- Broadcom Inc. Form 8-K (Google TPU long-term agreement and Anthropic capacity), April 6, 2026 · Broadcom (SEC filing) · Primary source · 2026-04-06 (retrieved 2026-11-07)
- [32]
Google named Citadel Securities as an organization choosing TPUs for its AI workloads. confirmedas of 2026-04-22
- Our eighth generation TPUs: two chips for the agentic era · Google (The Keyword) · Primary source · 2026-04-22 (retrieved 2026-11-07)
- [33]
In February 2026 The Information reported, according to SiliconANGLE, that Meta had agreed a multibillion-dollar deal to rent Google Cloud TPUs to train and run its models, while a separate deal to buy TPUs for its own data centers had not been reached. reportedas of 2026-02-26
- Google and Meta reportedly strike new, multibillion-dollar AI chip deal · SiliconANGLE · Secondary source · 2026-02-26 (retrieved 2026-11-07)
- Google and Meta reportedly strike new, multibillion-dollar AI chip deal · SiliconANGLE · Secondary source · 2026-02-26 (retrieved 2026-11-07)
- Google and Meta reportedly strike new, multibillion-dollar AI chip deal · SiliconANGLE · Secondary source · 2026-02-26 (retrieved 2026-11-07)
- Google and Meta reportedly strike new, multibillion-dollar AI chip deal · SiliconANGLE · Secondary source · 2026-02-26 (retrieved 2026-11-07)
- [34]
On November 25, 2025, Nvidia posted on X that it was "delighted by Google's success" and continues to supply Google, but that it is "a generation ahead of the industry" and offers greater performance, versatility and fungibility than ASICs designed for specific AI frameworks or functions. confirmedas of 2025-11-25
- Nvidia scoffs at threat from Google TPUs · The Register · Secondary source · 2025-11-25 (retrieved 2026-11-07)
- Nvidia says it's 'delighted' with Google's success, but says it is 'the only platform that runs every AI model' · Tom's Hardware · Secondary source · 2025-11-25 (retrieved 2026-11-07)
Revision history (1)
- Page created.
Created Nov 7, 2026. Last reviewed by an editor on Nov 8, 2026. Next scheduled review: Feb 7, 2027.
Cite this page
"Google TPU (Tensor Processing Unit)." ContentLora, updated Nov 7, 2026. https://contentlora.com/wiki/google-tpu
Spotted an error? Suggest a correction or emailcorrections@contentlora.com.