Next ’25 Infrastructure Innovations: Building the Foundation for AI at Scale

At Google Cloud Next ’25, several key infrastructure advancements were unveiled, designed to help businesses run AI and enterprise workloads with top-tier performance, global connectivity, and flexibility. Here’s a breakdown of the key updates.
Cloud WAN: Lightning-Fast Networking
Google Cloud’s new Cloud WAN is now available, offering:
-
Ultra-fast connections with near-zero latency for billions of users worldwide.
-
Up to 40% better network performance and 40% lower costs compared to traditional solutions.
This global network backbone connects data centers, branches, and campuses at "Google speed," ensuring fast and reliable connections.
AI Hypercomputer & Ironwood TPU: Powering AI at Scale
The AI Hypercomputer platform has received powerful updates to help businesses get the most out of their AI workloads:
-
Ironwood TPU (7th generation) pods offer 42.5 exaFLOPS, up to 4 times the performance of previous models.
-
vLLM for TPUs improves large-language model performance across platforms like Compute Engine and Vertex AI.
-
Pathways for Google Cloud provides a unified machine learning runtime for large-scale training and inference.
These upgrades mean faster model training, fewer bottlenecks, and better ROI on AI investments.
Workload-Optimized Storage & Compute: Supercharging Performance
New tools to support data-heavy and compute-intensive applications:
-
Rapid Storage (Preview): Zonal storage with <1 ms latency and 20x faster access than standard regional buckets.
-
Anywhere Cache (GA): A cache that reduces latency by up to 70%, keeping data close to GPUs/TPUs for faster processing.
-
Hyperdisk Exapools: Exabyte-scale block storage delivering TiB/s throughput for high-performance computing (HPC) and AI workloads.
These updates ensure data is processed faster and more efficiently, minimizing delays and maximizing throughput.
Google Distributed Cloud: AI On-Premises with Gemini
For businesses with specific data residency or low-latency needs, Gemini on Google Distributed Cloud (still in pre-announcement) offers:
-
On-premises AI running on NVIDIA DGX/HGX platforms with Gemini models.
-
Agentspace Search for secure, enterprise-level search and automation without compromising data residency.
This allows businesses to run powerful AI tools on-premises, ensuring compliance and smooth performance.
What It Means for Businesses
These innovations create the strong infrastructure foundation needed for next-generation AI and enterprise applications. With high-performance networking, powerful compute options, ultra-fast storage, and flexible on-premises solutions, businesses can scale their operations with confidence – whether it's for training complex models, running real-time AI services, or connecting global teams.
Contact us
© 2025 Terasky. All rights reserved