Google Unveils Virgo Network: A Megascale AI Data Center Fabric Connecting 134,000 Chips with 47 Petabits Per Second

Google unveils Virgo Network, a massive AI data center fabric connecting 134,000 chips with 47 petabits per second of bandwidth, delivering 4x faster speeds and 40% lower latency than its predecessor to power next-generation AI training at unprecedented scale.

• Google introduces Virgo Network, a megascale AI data center fabric built on a flat, two-layer non-blocking topology that connects up to 134,000 chips with 47 petabits per second of non-blocking bi-sectional bandwidth.
• Virgo Network delivers a 4x bandwidth increase per accelerator over the previous generation and achieves 40% lower unloaded fabric latency, addressing the explosive bandwidth demands and synchronized traffic bursts of modern AI training workloads.
• The architecture features independent switching planes for fault isolation, sub-millisecond telemetry for deep observability, and automated straggler and hang detection to maximize reliability and minimize recovery time at massive scale.