NVIDIA Vera Rubin NVL144

NVIDIA's next-generation rack-scale AI system pairing Vera CPUs with Rubin GPUs, detailed with a shipping timeline pulled forward to compete with AMD's newly announced Helion rack.

NVIDIA · AI hardware and silicon · infrastructure

Published in Feed

This entry is marked for review. Check the original sources before relying on its details.

Overview

NVIDIA detailed the shipping timeline for its Vera Rubin NVL144 rack-scale system on October 13, 2026, one day after AMD announced its competing Helion agentic-AI rack blueprint. The rack pairs Vera CPUs with 144 Rubin GPU dies (72 dual-die packages) over NVLink 6, which NVIDIA says delivers roughly 3.3x the inference throughput per rack of the prior-generation GB200 NVL72. NVIDIA pulled the shipping timeline forward from an original second-half-2027 target to cloud partners including CoreWeave, Oracle, and Microsoft Azure starting Q1 2027, and extended confidential computing across the full rack including CPU-GPU shared memory. A new NVLink Fusion option lets partners mix in custom ASICs alongside Rubin GPUs within the same rack. At GTC Berlin on October 21, 2026, NVIDIA described Vera Rubin as in full production and announced Google Cloud A5X instances built on Vera Rubin NVL72 with ConnectX-9 SuperNICs, with Google planning up to 960,000 Rubin GPUs across sites; coverage differs on whether A5X reaches general availability on Oct 21 or the following month, and this conflicts with the Q1 2027 cloud timeline above.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • 144 Rubin GPU dies (72 dual-die packages) paired with Vera CPUs over NVLink 6
  • Roughly 3.3x the inference throughput per rack of GB200 NVL72, per NVIDIA
  • Confidential computing extended across the full rack, including CPU-GPU shared memory
  • Shipping to CoreWeave, Oracle, and Microsoft Azure pulled forward to Q1 2027
  • NVLink Fusion option for mixing custom ASICs alongside Rubin GPUs in the same rack

Use cases

  • Hyperscale frontier-model training and inference
  • Cloud GPU rental fleets refreshing capacity ahead of 2027
  • AI labs needing higher inference throughput per rack at similar power envelopes
  • Partners building custom-silicon hybrid racks via NVLink Fusion

Pricing

Not sold direct to end customers; priced through OEM/cloud partner systems, list pricing undisclosed.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
13 Oct 2026
Record updated
22 Oct 2026
Record last checked
22 Oct 2026

Explore the Feed

Other entries filed under ai hardware and silicon; inclusion does not imply equivalent features.

  • Azure Maia 200 and Maia 300

    Microsoft says its Maia 200 AI accelerator is in production serving Copilot, and unveiled Maia 300 for 2027 deployment.

  • AMD Helion

    AMD's open rack-scale blueprint for agentic AI infrastructure, with CPU-only configurations up to 32,768 EPYC cores and an integrated model-call token router.

  • Structera X 3406 / 3506

    Marvell's second-generation CXL 3.2 memory expanders, the industry's first six-channel design, supporting up to 9TB of pooled memory per controller for AI inference.

Browse the Feed