Google Unveils Next-Gen AI Chips and Launches Agent Platform

A comprehensive AI toolkit for enterprises

At its annual Google Cloud Next conference in Las Vegas, Google rolled out a sweeping set of announcements: tools for building and managing AI agents, next-generation tensor processing units TPU 8t and TPU 8i, and a $750 million fund to drive adoption of agentic AI through consulting partners.

«Have a great #GoogleCloudNext! Watch select sessions and your favorite moments from the opening keynote on demand» — Google Cloud (@googlecloud), original post

The company's cloud division showcased a suite of solutions enabling businesses to build their own virtual assistants and monitor their performance. Among the new features is a dedicated inbox where AI agents submit reports on completed tasks.

The Gemini Enterprise Agent Platform gained two new modules — Memory Bank and Memory Profile — allowing agents to retain interaction history with individual users. A new Agent Simulation tool gives developers expanded testing capabilities before deploying agents in production environments.

Google also updated its Workspace suite and outlined a vision where AI agents fundamentally reshape the daily workflow of ordinary employees. The company stated that Gemini Enterprise will let workers build virtual assistants without writing any code.

Another addition is Projects, a collaborative workspace designed for interactions among employees and support operators. The tool aggregates information from multiple sources — including Workspace, Microsoft OneDrive, and corporate messaging apps — to provide unified context for work.

$750 million fund for consulting partners

Google Cloud announced a $750 million fund aimed at helping consulting firms such as McKinsey, Accenture, and Deloitte deploy agentic AI for their clients. DeepMind will grant select firms early access to Gemini AI models ahead of their public release.

The capital will be directed toward training engineers, building AI agents through Google's enterprise platform, and co-funding projects and pre-sales activities.

Kevin Ichhpurani, head of Google Cloud's global partner ecosystem, emphasized that consulting firms are at the center of the largest transformations happening on the client side and bring unique industry-specific expertise and business process knowledge.

TPU 8t and TPU 8i processors

The headline hardware announcement was a new generation of Google's custom tensor processing units, designed to accelerate AI workloads while reducing costs. The lineup features two variants:

• TPU 8t — optimized for AI model training and development;
• TPU 8i — tailored for inference workloads.

Google holds a strong position among companies designing proprietary AI chips, competing directly with Nvidia. Demand for TPUs has been rising sharply in Silicon Valley in recent months.

The new processors feature expanded on-chip memory, reducing response latency by eliminating the need to fetch data from external sources. TPU 8t units can be clustered in groups of up to 9,600. At that scale, power efficiency becomes paramount: TPU 8t delivers 124% more performance per watt compared to its predecessor, while TPU 8i achieves a 117% improvement.

Mark Lohmeyer, Google's vice president of compute infrastructure, noted that the core objective is achieving the lowest possible latency at the lowest possible cost per operation.

Why it matters

Google's announcements signal an intensifying race for dominance in enterprise AI infrastructure. The company is simultaneously advancing its hardware (TPU) and software ecosystem (agent platform, Workspace), creating a closed loop for business clients. The $750 million fund channeled through global consulting giants is designed to accelerate Google's technology penetration into the world's largest corporations.

The more-than-double improvement in power efficiency for the new TPU generation represents substantial engineering progress and could increase competitive pressure on Nvidia in the AI infrastructure segment.