Google Announces TPU v8 Chips and Gemini Enterprise Agent Platform at Cloud Next '26

Google Cloud held its annual Cloud Next ‘26 conference, where it declared the beginning of the “Agentic Era” and unveiled a suite of innovations for transforming businesses into agentic enterprises. The company unveiled eighth-generation Tensor Processing Units (TPUs) and launched the Gemini Enterprise Agent Platform, described as a comprehensive tool for building, scaling, governing and optimizing autonomous AI agents.

Google CEO Sundar Pichai congratulated the Google Cloud team for innovations over the last year.

' Google's models process 16 billion or more tokens per minute via direct API use by customers, up from 10 billion tokens per minute last quarter. The Gemini Enterprise Agent Platform includes an 'Agent Studio' for low-code development and 'Agent-to-Agent Orchestration' for AI agents to delegate tasks to one another.

Google introduced the TPU 8t chip, designed for high-speed model training and scalable up to 9,600 TPUs in a single superpod, offering three times the processing power of the previous generation and twice the performance per watt. The company also introduced the TPU 8i chip, optimized for real-time AI agent inference, delivering 80% better performance per dollar and engineered for near-zero latency.

Google announced it will offer Nvidia’s Vera Rubin NVL72 systems, providing customers flexibility in choosing architectures for their needs.

The announcements position AI to proactively complete complex business processes, moving beyond answering questions, according to details presented at the conference. Google emphasized a new approach to data and security in the Agentic Era, with the Agentic Data Cloud allowing agents to reason across a company’s data estate in real-time.

To protect autonomous systems, Google introduced Agentic Defense, combining threat intelligence with security measures to detect reasoning drift or suspicious agent behavior proactively.