Google chip and agent signals

Google is pairing new artificial-intelligence chips with new workplace agents, pushing the same Gemini strategy from data centers into office software. (blog.google) At Google Cloud Next on April 22, Google introduced two eighth-generation Tensor Processing Units, or TPUs: TPU 8t for training models and TPU 8i for running them in real time. Google said TPU 8i is built for autonomous agents that need to reason, plan and execute multi-step work quickly, while TPU 8t is designed to train large models in a single large memory pool. (blog.google) A TPU is Google’s in-house AI chip, the hardware that does the heavy math behind models like Gemini. In a technical deep dive, Google said the new systems are meant to handle long context windows, complex sequential logic and the split between training, post-training and live serving. (cloud.google.com) Google tied those chips to a broader “Agentic Enterprise” pitch at Next 2026. Chief Executive Thomas Kurian said the company is rolling out a Gemini Enterprise Agent Platform with tools including Agent Designer, long-running agents, Skills and Projects, alongside new storage and networking upgrades. (cloud.google.com) On the software side, Google Workspace has been moving the same idea closer to everyday office work. In December 2025, Google made Workspace Studio generally available, saying employees can build and share Gemini-powered agents inside Workspace without coding. (workspace.google.com) Google says those agents can automate tasks that used to rely on fixed rules, such as sorting messages, drafting content, prioritizing work and handing off multi-step processes between specialized agents. In its launch example, Kärcher used a chain of agents in Google Chat to evaluate feature ideas, check technical feasibility and draft a user story. (workspace.google.com) Colgate-Palmolive gives a sense of the scale Google is chasing in large companies. Google Workspace’s customer materials say Colgate migrated 28,000 users to Workspace in six months, while the company itself has about 38,000 employees across more than 130 offices worldwide. (workspace.google.com) Google Cloud said at Next that nearly 75% of Google Cloud customers now use its AI products, and that 330 customers processed more than 1 trillion tokens each over the last 12 months. Those figures suggest Google is trying to show customers that its AI business is no longer limited to model demos and pilot projects. (cloud.google.com) The open question is whether companies will trust agents with enough internal data and decision-making authority to make the hardware push pay off. Google’s answer this week was to present the stack as one system: Gemini models, Google chips, enterprise controls and agents embedded in the tools employees already use. (cloud.google.com; workspace.google.com)