# Virgo Network

> **TL;DR.** Virgo Network is Google's custom-built megascale AI data center fabric, announced April 22, 2026 at Google Cloud Next '26.

- **Category:** AI / Infrastructure / Networking
- **Stage:** validating
- **Age:** 55 days
- **Origin date:** 2026-04-22
- **First detected:** 2026-05-04
- **Canonical URL:** https://earlyterms.com/term/virgo-network
- **Sources:** 7 primary URLs

## Definition

Virgo Network is Google's custom-built megascale AI data center fabric, announced April 22, 2026 at Google Cloud Next '26. It replaces the multi-tier Jupiter Ethernet design with a flat, two-layer non-blocking topology built on high-radix switch ASICs — eliminating network hops that were adding latency and bandwidth bottlenecks to large-scale AI training runs.

The fabric can link [134,000 TPU 8t chips](https://cloud.google.com/blog/products/networking/introducing-virgo-megascale-data-center-fabric) in a single data center, delivering 47 petabits per second of bisectional bandwidth with 40% lower unloaded latency than the prior generation. Combined with optical circuit switches (OCS), Virgo scales to over one million TPUs in a single logical training cluster — a capability Google Engineering Fellow Arjun Singh called a "generational" leap enabling the company's "campus-as-a-computer" design philosophy.

## Analogy

Think of Virgo as replacing the highway interchange with a direct overpass — fewer turns, higher throughput.

## Why it's emerging now

Google unveiled Virgo Network on April 22, 2026 as a core component of its TPU 8t infrastructure — the first networking design capable of linking over one million AI accelerators in a single logical cluster. The announcement landed at Cloud Next '26 alongside TPU 8t and 8i, positioning Google's full-stack AI infrastructure against Nvidia's NVLink and the broader InfiniBand ecosystem.

## Related terms

- *parent:* AI Hypercomputer
- *related:* TPU 8t
- *competitor:* InfiniBand
- *competitor:* NVLink
- *related:* RoCE
- *related:* Jupiter Network
- *competitor:* DGX Spark
- *related:* ai-capex
- *related:* ai-supply

## Sources

1. [Google Cloud Blog — Introducing Virgo Network megascale data center fabric (Apr 22, 2026)](https://cloud.google.com/blog/products/networking/introducing-virgo-megascale-data-center-fabric)
2. [Google Cloud Blog — TPU 8t and TPU 8i technical deep dive (Apr 22, 2026)](https://cloud.google.com/blog/products/compute/tpu-8t-and-tpu-8i-technical-deep-dive)
3. [Google Blog — Our eighth generation TPUs: two chips for the agentic era (Apr 22, 2026)](https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/)
4. [The Register — Google dual tracks TPU 8 to conquer training and inference (Apr 22, 2026)](https://www.theregister.com/2026/04/22/google_tpu8_dual_track_training_inference/)
5. [The Next Platform — New Google Networks Tuned Up For GenAI Inference And Training (Apr 28, 2026)](https://www.nextplatform.com/connect/2026/04/28/new-google-networks-tuned-up-for-genai-inference-and-training/5218978)
6. [SiliconANGLE — Google rolls innovations in mega-scale networking for the agentic era (Apr 22, 2026)](https://siliconangle.com/2026/04/22/google-rolls-innovations-mega-scale-networking-agentic-era/)
7. [Hacker News — Our eighth generation TPUs: two chips for the agentic era (454 pts, Apr 22, 2026)](https://news.ycombinator.com/item?id=47862497)

---
_Generated by EarlyTerms · https://earlyterms.com/term/virgo-network_