Thursday, October 22, 2026
AI 인프라 · 뉴스 & 분석
홈 › 데이터센터 › 리포트
데이터센터 · 리포트

Google Cloud는 Nvidia Vera Rubin 인프라를 96만 개 GPU 규모로 확장할 계획이다.

이 규모의 배치는 Vera Rubin 공급을 빠듯하게 만들고 GPU 클라우드 및 네오클라우드의 경쟁 기준을 높이는 대규모 용량 확충이 될 것이다.
업계 전문지Slicast · 2026년 10월 21일 13:15 UTC · 미국 · 출처: Crypto Briefing
중요도 90

Google Cloud plans to expand its NVIDIA Vera Rubin infrastructure to as many as 960,000 GPUs. The expansion will be delivered through a new family of A5X instances built on NVIDIA's Vera Rubin NVL72 systems.

**What Google actually built**

The A5X instances are bare-metal, giving customers direct access to the physical hardware without a virtualization layer between their software and the chips.

Each instance is powered by an NVIDIA Vera Rubin NVL72 rack. A single rack combines 72 Rubin GPUs and 36 Vera CPUs, all linked together with NVLink 6.

Scaling beyond a single rack requires additional networking. Google is using its Virgo networking fabric and NVIDIA ConnectX-9 SuperNICs to link systems across multiple sites.

**The performance claims**

The headline numbers are ambitious. The A5X instances reportedly deliver up to 10 times lower inference cost per token than the previous generation and are claimed to offer up to 10 times higher throughput per megawatt.

**The timeline so far**

NVIDIA launched the Vera Rubin platform on January 5, 2026. Google then announced the A5X instances at Google Cloud Next in April 2026. The first A5X instances went live on October 21, 2026, and general availability is planned for November 2026.

The first customer is Ineffable Intelligence, a London-based company that is using A5X to run complex reinforcement learning systems.

Google has also signaled a planned European expansion of the infrastructure. The announcement materials do not specify sites or dates for that rollout.

**What this means for the cloud and chip landscape**

There are real risks to track. The 960,000 GPU figure is a stated capacity ceiling across multisite deployments, not a count of chips already running.

The next markers to watch are the November 2026 general availability date and any concrete details on the European expansion.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

원문 보기
Google Cloud는 Nvidia Vera Rubin 인프라를 96만 개 GPU… · Slicast