How we evolved Google’s global and data center networks for the AI era

Google has evolved its network infrastructure to support AI workloads, moving from the cloud era to the AI era with fundamentally different requirements.

• •Virgo Network, a new data center fabric engineered for AI, connects up to 134,000 TPU 8t chips with 47 petabits/sec non-blocking bandwidth and 40% lower latency than previous generations
• •Autonomous reliability capabilities enable efficient fault detection and isolation with minimal manual intervention, maximizing workload goodput
• •AI-native Cloud Interconnect provides flexible high-bandwidth connectivity up to 3.2 Tbps, reducing AI training data transfer time from 22.2 hours to 0.7 hours
• •The global network spans over 10 million kilometers of fiber, connects 43 cloud regions with 200+ edge locations, ensuring low latency for AI inference at global scale