AI-FRONTIERApril 23, 20261 min read6 views

Here’s how our TPUs power increasingly demanding AI workloads.

Google's custom-designed Tensor Processing Units (TPUs) are the backbone of its AI infrastructure, offering massive scale for complex mathematical computations. The latest generation delivers 121 exaflops of power, doubling the bandwidth of previous versions to meet modern AI demands.

Behind the Google products you use every day are custom chips designed for one job: doing math at massive scale. They're called TPUs, or Tensor Processing Units.

We designed TPUs from the ground up more than a decade ago specifically to run AI models. Basically, it takes a lot of math for AI models to work, and TPUs can do complex math super quickly: The newest generation of TPUs can process 121 exaflops of compute power with double the bandwidth of previous generations.

Learn more about these tiny but mighty processors in the video below.

Source: Google AI Blog

More in this category

ai-frontier

The apps, gadgets, and tools every reader needs

This week's guide highlights the ultimate tech setup for avid readers, featuring top-rated e-readers, note-taking apps, and digital tools. Plus, discover exciting new software updates, gadgets, and media recommendations.

ai-frontier

Your Period Tracker Is (Probably) Spying on You

A recent Mozilla audit reveals that popular period tracking apps are sharing highly sensitive user data with third parties, highlighting broader privacy and security concerns alongside major global cyber incidents this week.

ai-frontier

How Google’s New Gemini Rates Work and How to Track Your Usage

Earlier this summer, Google rolled out upgrades to Gemini but also introduced new usage limits based on computing power. Here is how the new system works and how you can track your usage.

ai-frontier

Prompt Injection Attacks Are Thwarting AI Hacking Agents

Defenders are turning the tables on AI hacking agents by using prompt injection as a defense mechanism. Dubbed "context bombing," the technique plants forbidden commands in decoy data to trigger safety guardrails and shut down attacking LLMs.

ai-frontier

TikTok is testing an AI likeness detection tool

TikTok is testing an opt-in tool that scans for unauthorized AI-generated likenesses of creators, allowing them to report deepfakes and protect their digital identity.

ai-frontier

Apple’s plot to crush OpenAI

Apple has filed a lawsuit against OpenAI, sparking intense debate over whether the tech giant is trying to eliminate a rising competitor or simply capitalizing on OpenAI's current vulnerabilities.