AI Infrastructure DatabaseThe chips, power and companies behind AI
Saved
Menu
Home How It's Built Watchlist The Stack
Hardware ChipsCompare Chips Servers & SystemsNetworking MemoryFoundries & Packaging Cooling
Facilities Data CentersPower ProjectsCloud & GPU Compute
Intelligence AI ModelsSoftware AI ProductsBenchmarks Research PapersDatasets Concepts
Ecosystem CompaniesStocks PeopleTimeline TrendsFuture of AI
Learn News Ask the Data AI Agents
About About us Methodology Data & API Disclaimer
Options
DATA & API

Free read-only JSON access to the site's cached data.

Dark mode

Guided View
New to chips, data centers and model jargon? We explain key terms as you browse, in plain English. Same data, with the help built in.

Expert View
You already know the stack. Just the data: clean, fast and compact, with no extra explanations. This is the default view.

Interface language
‹ Back
Home › Servers & Systems › NVIDIA Vera Rubin NVL72

NVIDIA Vera Rubin NVL72

System Rack-Scale System Shipping

The third-generation NVLink rack: 72 Rubin GPUs and 36 Vera CPUs in the Oberon frame.

You're looking at a system record: an assembled machine, a server or a whole rack, with what's inside it and where it's deployed.

Vera Rubin NVL72 is NVIDIA's 2026 rack-scale system and the successor to GB300 NVL72. It keeps the Oberon rack layout of 72 GPU sockets and 36 CPU sockets, filled with Rubin GPUs, Vera CPUs, NVLink 6 switches, ConnectX-9 SuperNICs and BlueField-4 DPUs. NVIDIA's spec table rates the rack at 3,600 PFLOPS of NVFP4 inference (sparse) with 20.7 TB of HBM4. First styled NVL144, counting GPU dies, it was restyled NVL72 to count packages. Production shipments began in August 2026; CoreWeave and Oracle Cloud made it available in September and October 2026.

Specifications & key facts

Each figure is labeled by basis: vendor spec, vendor claim, reported, estimate, or official. "Not publicly disclosed" means exactly that: we never invent values.

Overview

Configuration 72 Rubin GPUs + 36 Vera CPUs (36 Vera Rubin superchips)Vendor spec
NVLink 6 domain All 72 GPUs at 3 TB/s each: 216 TB/s aggregateVendor spec NVIDIA spec table (retrieved 16 Nov 2026); the January 2026 launch release gave 3.6 TB/s per GPU and 260 TB/s per rack
CPU cores 3,168 Olympus cores, 6,336 threads per rackVendor spec
Naming Styled NVL144 at GTC 2025 (die count); NVL72 from 2026 (package count)Reported Next Platform, Mar 2026
Availability Production shipments from Aug 2026; available on CoreWeave 30 Sep 2026; GA on OCI 27 Oct 2026Official

Memory

HBM4 20.7 TB per rack; 1,400 TB/sVendor spec NVIDIA spec table (retrieved 16 Nov 2026)
CPU memory up to 54 TB LPDDR5XVendor spec

Compute

NVFP4 inference 3,600 PFLOPSVendor spec Sparse, rack
NVFP4 training 2,520 PFLOPSVendor spec Dense, rack
FP8/FP6 training 1,260 PFLOPSVendor spec Dense, rack

Power

Rack power Not publicly disclosed

How this connects to the rest of the stack

This record's links at a glance: click a group to jump to it. +1 more groups below.

Designed by

Memory technology

Interconnect & networking

Available from

Successor to

Sources

NVIDIA Vera Rubin NVL72 specifications ↗ · NVIDIA Rubin platform launch (CES, Jan 2026) ↗ · NVIDIA and Oracle: Vera Rubin GA on OCI (Oct 2026) ↗ · CoreWeave: Vera Rubin NVL72 in production (Sep 2026) ↗ · Next Platform: GTC 2026 roadmap (Mar 2026) ↗

Record reviewed as of 2026-11-16; see Methodology.