Google Cloud Strengthens Agentic AI with Gemini and Advanced TPUs

At the Google Cloud Next conference, Google announced a series of significant updates to enhance the use of agentic AI in businesses. This initiative aims to automate the business processes of its clients, and currently, 75% of Google Cloud customers are already using AI in their operations. This adoption is not surprising, given the integration of AI into flagship Google products such as Docs, Sheets, and Gmail. Google continues its vision of the "agentic enterprise," where AI agents play a central role in automating tasks.

Agentic AI is distinguished by its ability to enable agents, or bots, to perform tasks autonomously, requiring minimal human supervision. This technology is transforming AI, particularly in the areas of coding and administrative tasks. Tech companies are betting on tools like OpenClaw, Claude Code, and OpenAI's Codex to fulfill the promise of AI in automating large swathes of tasks. Google is giving its enterprise technology the same agentic transformation.

Thomas Kurian, CEO of Google Cloud, indicated that agentic AI is the direction envisioned for the future of AI technology. This year's updates focus on the security of AI processes, their connection to internal systems, and the optimization of performance, scale, and costs associated with the operation of agents. Kurian emphasized the importance of these aspects to ensure successful adoption of agentic AI by businesses.

The Gemini enterprise agent platform is at the heart of this transformation. It offers businesses the ability to oversee and create AI agents through a dedicated application. Gemini includes a new agent design tool, which allows users to program tasks to be executed across different applications, thereby facilitating the integration of AI into existing processes.

Meanwhile, Google unveiled two new eighth-generation TPUs: the 8T and 8I. These tensor processing units are not just simple processors; they are specifically designed for tech companies engaged in compute-intensive tasks, such as AI development. Google claims that the 8T chip is optimized to make training more efficient, with processing power three times that of the seventh-generation Ironwood. The 8I chip, on the other hand, is dedicated to inference, featuring an 80% improvement in available memory thanks to SRAM, and it integrates approximately 11,152 chips into a single system, marking a significant advancement for enterprise processing capabilities.