3 things to know about Ironwood, our latest TPU

Recorded: Nov. 26, 2025, 1:03 a.m.

Original

Summarized

3 things to know about Ironwood, Google's latest TPU

The Keyword

3 things to know about Ironwood, our latest TPU

Twitter

Facebook

Mail

Copy link

Home

Product news

Android, Chrome & Play

Android

Chrome

Chromebooks

Google Play

Wear OS

See all

Platforms & Devices

Fitbit

Google Nest

Pixel

See all

Explore & Get Answers

Gemini

Maps

News

Shopping

See all

Connect & Communicate

Classroom

Photos

Registry

Translate

In the Cloud

Google Workspace

More on the Cloud Blog

Google Cloud

See all

See all product updates

Company news

Outreach & initiatives

Arts & Culture

Education

Entrepreneurs

Public Policy

Sustainability

See all

Technology

Developers

Health

Google DeepMind

Google Labs

Safety and security

See all

Inside Google

Data centers and infrastructure

Doodles

Googlers

Life at Google

See all

Around the globe

Google in Asia

Google in Europe

Google in Latin America

See all

Authors

Sundar Pichai, CEO

Demis Hassabis, CEO and Co-Founder, Google DeepMind

Kent Walker, SVP

James Manyika, SVP

Ruth Porat, President & Chief Investment Officer

See all

Feed

Press corner

RSS feed

Breadcrumb

Products

Google Cloud

3 things to know about Ironwood, our latest TPU

Nov 25, 2025
·

Twitter

Facebook

Mail

Copy link

Our seventh-gen Tensor Processing Unit is here! Learn what makes Ironwood our most powerful and energy-efficient custom silicon to date.

Ari Marini

Keyword Contributor

Read AI-generated summary

General summary

Ironwood our seventh-generation Tensor Processing Unit (TPU) is now available for Cloud customers. It's custom built for high-volume low-latency AI inference and model serving. Ironwood can scale up to 9216 chips in a superpod which helps to reduce the compute-hours and energy required for training and running cutting-edge AI services.

Summaries were generated by Google AI. Generative AI is experimental.

Twitter

Facebook

Mail

Copy link

Today's most advanced AI models, like those powering complex thinking and calculations, need speed and efficiency from the hardware that powers them. That's why at Cloud Next in April, we unveiled Ironwood, our seventh-generation Tensor Processing Unit (TPU).

Ironwood is our most powerful, capable, and energy-efficient TPU yet, designed to power thinking, inferential AI models at scale.

By acting as a hugely efficient parallel processor, Ironwood excels at managing massive calculations and significantly minimizes the internal time required for data to shuttle across the chip. This breakthrough dramatically speeds up complex AI, making models run significantly faster and smoother across our cloud.And now, Ironwood is here for Cloud customers.Here are three things to know about it.1. It’s purpose-built for the age of inferenceAs the industry’s focus shifts from training frontier models to powering useful, responsive interactions with them, Ironwood provides the essential hardware. It’s custom built for high-volume, low-latency AI inference and model serving. It offers more than 4X better performance per chip for both training and inference workloads compared to our last generation, making Ironwood our most powerful and energy-efficient custom silicon to date.

2. It’s a giant network of powerTPUs are a key component of AI Hypercomputer, our integrated supercomputing system designed to boost system-level performance and efficiency across compute, networking, storage and software. At its core, the system groups individual TPUs into interconnected units called pods. With Ironwood, we can scale up to 9,216 chips in a superpod. These chips are linked via a breakthrough Inter-Chip Interconnect (ICI) network operating at 9.6 Tb/s.

Part of an Ironwood superpod, directly connecting 9,216 Ironwood TPUs in a single domain.

This massive connectivity allows thousands of chips to rapidly communicate and access a staggering 1.77 Petabytes of shared High Bandwidth Memory (HBM), overcoming data bottlenecks for even the most demanding models. This efficiency significantly reduces the compute-hours and energy required for training and running cutting-edge AI services.3. It’s designed for AI with AIIronwood is the result of a continuous loop at Google where researchers influence hardware design, and hardware accelerates research. While competitors rely on external vendors, when Google DeepMind needs a specific architectural advancement for a model like Gemini, they collaborate directly with their TPU engineer counterparts. As a result, our models are trained on the newest TPU generations, often seeing significant speedups over previous hardware. Our researchers even use AI to design the next chip generation — a method called AlphaChip — which has used reinforcement learning to generate superior layouts for the last three TPU generations, including Ironwood.

POSTED IN:

Google Cloud

Related stories

Google in Asia
Here’s how researchers in Asia-Pacific are using AlphaFold

By

Rachel Teo

Nov 26, 2025

AI
Get an in-depth look at Gemini 3 with CEO Sundar Pichai.

Nov 25, 2025

Google Labs
4 ways to refine your content in Flow

By

Anika Ahluwalia

Nov 25, 2025

AI
16 Google AI tips for stress-free holiday hosting in 2025

By

Molly McHugh-Johnson

Nov 21, 2025

Learning & Education
Develop a deeper understanding with interactive images in Gemini.

By

Dave Messer

Nov 20, 2025

AI
How we’re bringing AI image verification to the Gemini app

By

Pushmeet Kohli

&

Laurie Richardson

Nov 20, 2025

Jump to position 1

Jump to position 2

Jump to position 3

Jump to position 4

Jump to position 5

Jump to position 6

Let’s stay in touch. Get the latest news from Google in your inbox.

Subscribe
No thanks

Privacy

Terms

About Google

Google Products

About the Keyword

Help

Global (English)

Africa (English)

Australia (English)

Brasil (Português)

Canada (English)

Canada (Français)

Česko (Čeština)

Deutschland (Deutsch)

España (Español)

France (Français)

India (English)

Indonesia (Bahasa Indonesia)

Italia (Italiano)

日本 (日本語)

대한민국 (한국어)

Latinoamérica (Español)

الشرق الأوسط وشمال أفريقيا (اللغة العربية)

MENA (English)

Nederlands (Nederland)

Polska (Polski)

Portugal (Português)

ประเทศไทย (ไทย)

Türkiye (Türkçe)

台灣 (中文)

Ironwood, Google’s seventh-generation Tensor Processing Unit (TPU), represents a significant advancement in AI hardware designed to accelerate inference workloads. This new TPU is built for high-volume, low-latency AI inference and model serving, providing a crucial boost to Google’s AI capabilities. Here’s a breakdown of three key aspects of Ironwood:

1. **Designed for Inference at Scale:** Ironwood is specifically engineered for the shift in the AI landscape – moving beyond training complex models to deploying and utilizing them for real-time interactions. Unlike previous generations, it's optimized for inference, boosting efficiency for deploying and running AI services. Achieving up to 4x better performance per chip for both training and inference, it significantly improves speed and responsiveness.

2. **Massive Interconnectivity:** Ironwood leverages Google’s AI Hypercomputer, a system comprised of interconnected TPUs called pods. A superpod containing 9,216 Ironwood chips is facilitated by a breakthrough Inter-Chip Interconnect (ICI) network operating at 9.6 Tb/s. This immense connectivity enables thousands of chips to rapidly communicate and access a substantial 1.77 Petabytes of Shared High Bandwidth Memory (HBM), effectively mitigating data bottlenecks within the most demanding models.

3. **Research-Driven Innovation:** A key differentiator for Ironwood is its close collaboration between Google DeepMind researchers and TPU engineers. This integrated approach, enabled by AlphaChip – a method utilizing reinforcement learning to design next-generation chip layouts – ensures that the hardware aligns precisely with the needs of cutting-edge AI models. This ongoing feedback loop translates to optimized performance and a continuous stream of improvements for Google’s AI systems.