Google Cloud Next ‘26: Vertex AI rebranded as Gemini Enterprise Agent Platform in major infrastructure overhaul
- Marijan Hassan - Tech Journalist
- 32 minutes ago
- 3 min read
Google Cloud has officially entered the "Agentic Era," rebranding its flagship Vertex AI platform to the Gemini Enterprise Agent Platform and unveiling eighth-generation TPUs to power a new wave of autonomous business agents. During the Google Cloud Next ‘26 conference in Las Vegas, CEO Thomas Kurian declared that the industry has moved beyond simple chatbots toward "agentic workflows."
The three-day event was dominated by the introduction of a unified, end-to-end control plane for building, securing, and orchestrating millions of AI agents that can perform long-running business tasks without human intervention.
The Evolution: Gemini Enterprise Agent Platform
The most significant strategic move was the consolidation of all AI development tools under the Gemini Enterprise Agent Platform brand. This replaces the Vertex AI roadmap, signaling that Google is no longer selling a collection of tools, but a complete "operating system" for AI.
Agent Studio: A new low-code interface that allows business users to build and test agents using natural language.
A2A Protocol (Agent-to-Agent): Google announced that its Agent-to-Agent protocol, now governed by the Linux Foundation, has reached version 1.2. This allows a Google agent to seamlessly hand off tasks to agents built on Salesforce or AWS.
Model Garden Expansion: The platform now provides access to over 200 models, including the newly debuted Gemini 3.1 Pro and Gemini 3.1 Flash Image (Nano Banana 2), alongside third-party models like Anthropic’s Claude 4.7.
Hardware: The dual-chip TPU 8 strategy
To support the massive compute requirements of these agents, Google unveiled its eighth-generation Tensor Processing Units, opting for a specialized dual-chip architecture:
TPU 8t (Training): Optimized for compute-intensive model training, offering nearly 3x the performance per pod compared to previous generations.
TPU 8i (Inference): A dedicated chip for serving models, designed for near-zero latency. Google claims the 8i delivers 80% better performance-per-dollar than its predecessors.
Virgo Network & Managed Lustre: A new data center fabric and storage system capable of moving 10 terabytes of data per second, ensuring that data bottlenecks do not slow down agentic processing.
The Agentic Data Cloud
Google also introduced the Agentic Data Cloud, a new architecture designed to give AI agents a "real-time memory" of an entire enterprise.
Knowledge catalog: An autonomous system that tags and maps a company’s entire data estate, allowing agents to understand internal lingo and context.
Cross-cloud lakehouse: Standardized on Apache Iceberg, this allows agents to query data residing in AWS or Azure without the need for expensive and slow data migration.
Security and Governance: Agentic SecOps
Recognizing the risks of autonomous agents, Google launched Agentic SecOps, a suite of security tools built for the AI era:
Threat hunting agent: Proactively hunts for novel attack patterns using Google’s massive threat intelligence database.
Model Armor: A specialized defense layer designed to block "indirect prompt injection", where an agent is tricked by malicious instructions hidden in a document it is processing.
FedRAMP high authorization: The platform has achieved the highest level of US government security certification, allowing it to be used in sensitive public sector environments.
The "75% AI-Generated" Milestone
During his keynote, Sundar Pichai revealed a staggering statistic: 75% of all new code written at Google is now AI-generated, up from 25% just a year ago. This serves as the primary "proof of concept" for the company’s push toward an agent-driven economy, proving that AI-led development is no longer a pilot program, but a production-grade reality.
