/// Reserve your capacity

Register your
interest.

The NVIDIA Vera Rubin NVL72 is coming to our UK sovereign region in 2027. Register your interest now to be among the first to secure capacity.

NVIDIA - Vera Rubin NVL72

Design visualisation - not final CAD

Inside the NVL72

/// Scale

Facility: UK edge, invite only
Server hall: Aisles of NVL72
Rack: 600 x 1,200 x 2,300 mm
Column: 27 trays, 27U of continuous pitch
Compute tray: 1U, front-accessible NVMe
Superchip module: 1 Vera CPU, 2 Rubin GPUs

/// Platform

Compute trays: 18
NVLink switch trays: 9
Power shelves: 4 x 3RU, 110 kW each
Busbar: 52 V DC, 5,000 A
Storage: 4 x E1.S Gen6 + 1 x E1.S Gen5 boot
Cooling: 100% direct liquid

An NVIDIA Vera Rubin NVL72 rack, front door open on the tray column

Step inside, pull a tray out, take a module apart, right down to a reticle-sized die.

Open the digital twin

Generation over generation

NVFP4 inference per rack

Rubin

3,600 PFLOPS

Blackwell

1,440 PFLOPS

Memory bandwidth per rack

Rubin

1,580 TB/s

Blackwell

576 TB/s

NVLink per GPU

Rubin

3.6 TB/s

Blackwell

1.8 TB/s

Hosted in the UK. Liquid cooled.

A researcher at a microscope beside a screen modelling a protein structure

/// Sovereignty

A UK home for your data

Secure, compliant infrastructure, built for sovereign AI, hosted entirely in the UK, with the compliance and residency guarantees that regulated sectors depend on.

A compute tray slid out of a liquid-cooled NVL72 rack, its cold plates and modules exposed

/// Cooling

Every watt leaves as water

Direct-to-chip cold plates on every GPU and CPU, dripless blind-mate connectors, and a loop that stays charged even while a tray is pulled.

Where next

/// Reserve your capacity