Google Cloud unveils Gemini agent platform & new TPUs

Google Cloud has introduced a Gemini Enterprise Agent Platform, new cybersecurity products and eighth-generation Tensor Processing Units as part of a broader push to expand its AI business, Chief Executive Officer Sundar Pichai said.

The announcements span software, security and infrastructure as Google seeks to meet rising demand from companies building and managing AI agents. Google's first-party models now process more than 16 billion tokens per minute through direct customer API use, up from 10 billion in the previous quarter, Pichai said.

He added that in 2026, just over half of Google's overall machine learning compute investment is expected to go to the Cloud business, supporting customers and partners as demand for AI services grows.

Agent platform

The centrepiece of the update is the new Gemini Enterprise Agent Platform, which Google Cloud said will help organisations build, scale, govern and manage large numbers of AI agents. It follows the earlier launch of Gemini Enterprise, which Pichai described as an end-to-end system for the agentic era.

Usage is growing: paid monthly active users rose 40% quarter on quarter in the first quarter, according to Pichai.

The figures suggest the market is moving beyond experimentation. Companies are no longer just asking whether they can build an agent, but how to control and oversee many of them across business functions.

"That's why we're introducing our new Gemini Enterprise Agent Platform. It provides the secure, full-stack connective tissue you need to build, scale, govern and optimize your agents with confidence - a mission control for the agentic enterprise," Pichai said.

Security focus

Google Cloud also expanded its AI security portfolio, releasing a set of new agentic tools for threat detection. Pichai said they are part of a cybersecurity platform that combines Google's Threat Intelligence and Security Operations products with Wiz's Cloud and AI Security Platform.

Google is also launching Wiz's AI Application Protection Platform, or AI-APP, designed to provide protection from code to cloud to runtime across multicloud, hybrid and AI environments.

The security push reflects a wider industry shift, as providers try to show that AI can be used not only to automate attacks but also to defend against them. Google is making the case for the latter, arguing that customers can use AI systems to improve detection and response.

To support that claim, Pichai said Google's Security Operations Centre agents now automatically triage tens of thousands of unstructured threat reports each month and have reduced threat mitigation time by more than 90%.

He added that Google is also using Gemini-based AI agents, including a system called CodeMender, to identify and fix critical software flaws.

TPU update

On the infrastructure side, Google introduced its eighth-generation TPU line with two versions aimed at different workloads. TPU 8t is intended for training, while TPU 8i is designed for inference.

TPU 8t can scale to 9,600 TPUs and 2 petabytes of shared high-bandwidth memory in a single superpod, according to Pichai. He said it offers three times the processing power of Ironwood and up to twice the performance per watt.

TPU 8i, meanwhile, connects 1,152 TPUs in a single pod. Google said the design is meant to reduce latency and increase throughput when running large numbers of AI agents at the same time.

The chips will be offered to cloud customers alongside NVIDIA GPU instances, putting Google in direct competition with other cloud providers that are pairing proprietary silicon with third-party accelerators to attract AI workloads.

Customer zero

Pichai said Google continues to use its own products internally before taking them to market, referring to the company as "customer zero" for its technology. He cited coding, security and operations as recent examples.

In software development, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last autumn, he said. The company is also moving towards what he described as agentic workflows, in which engineers coordinate autonomous digital teams to complete tasks.

One internal code migration project completed by agents and engineers together was finished six times faster than would have been possible a year earlier with engineers alone, Pichai said. He also said the initial release of the Gemini app on MacOS was built with Google's agentic development platform, Antigravity, moving from idea to a native Swift app prototype in a few days.

Marketing was another area where Google highlighted internal AI use. Teams working on the launch of Gemini in Chrome used the company's models to generate thousands of versions of creative assets, cutting turnaround time by 70% and increasing conversions by 20%, Pichai said.

"Today, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall. We're now shifting to truly agentic workflows. Our engineers are orchestrating fully autonomous digital task forces, firing off agents and accomplishing incredible things," Pichai said.