NVIDIA used GTC Berlin on Oct. 21 to announce Vera Rubin cloud instances, a new reference humanoid robot and European cloud expansions. Only some of it is usable today: Isaac Lab 3.0 and a robotics asset library are available now, Google Cloud's Vera Rubin instances come next month, and the humanoid arrives in early 2027.

What you can get, and when

  • Now: Isaac Lab 3.0, an open-source framework for training robot policies, is generally available. The SimReady Central Library, with more than 2,000 simulation-ready robot and factory assets, and the Human 1M Dataset are on Hugging Face (NVIDIA).
  • November 2026: Google Cloud's A5X instances, built on Vera Rubin NVL72 systems, reach general availability. NVIDIA's post says "coming next month" and gives no price or region list.
  • Early 2027: the Agile ONE SN reference humanoid.
  • Undated: Isaac OS, and the open reference workflows for Isaac GR00T 1.7, Cosmos and dexterous manipulation.

The Vera Rubin cloud

A5X pairs Vera Rubin NVL72 with ConnectX-9 SuperNICs and Google Cloud's Virgo Network. Google Cloud plans to scale it to as many as 960,000 Rubin GPUs across multiple sites, with expansion to Europe planned.

London-based Ineffable Intelligence is the first customer on A5X, for large-scale reinforcement learning. WPP ran a creative-agent reference design that, in initial tests, fit on one A5X instance instead of a multi-node cluster.

Confidential computing is also generally available on Google Cloud's G4 VMs, which use RTX PRO 6000 Blackwell Server Edition GPUs.

The reference humanoid

Agile ONE SN is built by Agile Robots on Isaac GR00T and runs on Jetson Thor. Stanford University and ETH Zurich are named as interested institutions.

  • Size: 1.40 m (4.59 ft) tall, 40 kg (88 lb)
  • Body: 31 degrees of freedom, 3 kg payload per arm
  • Hands: two five-finger WUJI Hand 2 units, 20 degrees of freedom each, with tactile sensing
  • Compute: Jetson Thor, 2,070 FP4 teraflops, 128 GB unified memory
  • Availability: early 2027; no price given

Europe's cloud partners

NVIDIA's partner roundup says Nscale and Verda plan to deploy Vera Rubin and Nebius will pair Groq 3 LPX with Vera Rubin NVL72, without dates. Mistral's Vera Rubin capacity is planned for delivery in 2027. Crusoe is developing liquid-cooled modular sites for Vera Rubin inference.

Why it matters

For most readers the near-term change is software: Isaac Lab 3.0 and the asset library can be used today, while the Vera Rubin instances arrive next month on Google Cloud and the humanoid in early 2027. The humanoid is a reference design meant for developers, so it is a research and pilot platform, not a product to buy for a home. European Vera Rubin capacity mostly has no date yet.

For related coverage, see Nvidia's reported RTX Pro 5500 China orders and the reported pause in RTX 5090 production.