Google Unveils Expanded AI Infrastructure for the Agentic Era at Cloud Next
Google introduces a comprehensive expansion of its AI Hypercomputer portfolio, including new TPUs, NVIDIA-powered instances, and advanced networking to support complex agentic AI workloads.
Google announces a major expansion of its AI Hypercomputer portfolio at Cloud Next to support the evolving demands of agentic AI. Unlike traditional chat interfaces, agentic AI relies on a primary agent that breaks down user goals into specific tasks for a fleet of specialized agents that collaborate in real-time. This complex chain reaction requires a unified infrastructure stack that combines purpose-built hardware, open software, and flexible consumption models to prevent spiraling costs and performance bottlenecks.
The updated infrastructure lineup features Google's eighth-generation TPU 8t and TPU 8i chips, which form the same foundation that powers the flagship Gemini models. Additionally, the company introduces A5X bare metal instances powered by NVIDIA's Vera Rubin NVL72 systems and new Axion N4A virtual machines equipped with custom Arm-based CPUs. These hardware upgrades work alongside fourth-generation Intel and AMD x86-based VMs to provide enterprises with diverse processing options for their specific AI workloads.
To support this massive computational power, Google unveils the Virgo Network, a breakthrough data center fabric designed specifically for AI workloads. The company also adds high-performance storage and data management solutions like Google Cloud Managed Lustre, Z4M VMs with RDMA capabilities, and a dedicated KV Cache storage subsystem. Finally, native PyTorch support for TPUs and new Google Kubernetes Engine capabilities ensure that developers easily deploy and manage these advanced agentic systems at scale.