LEVIATHAN SYSTEMS
NEXT GENERATION

NVIDIA Rubin Infrastructure & Deployment_

Rubin is the platform after Blackwell — pairing the new Rubin GPU with NVIDIA's Vera CPU, HBM4 memory, and NVLink 6. The teams already deploying Blackwell today are the ones who will deploy Rubin first. Leviathan is planning power, cooling, and fabric for Rubin now.

What Is NVIDIA Rubin?_

Rubin is NVIDIA's next data center GPU architecture, following Blackwell/Blackwell Ultra. The flagship rack system, Vera Rubin NVL144, combines Rubin GPUs with the custom Arm-based Vera CPU and moves to HBM4 memory and NVLink 6 — a large jump in interconnect bandwidth and per-rack compute.

It is followed by Rubin Ultra NVL576, which pushes toward ~600 kW racks. The practical implication for operators: power delivery (800V DC), liquid-cooling capacity, and back-end fabric all need to be planned a generation ahead. Retrofitting later is far costlier than designing for it now.

Rubin Platform Specifications_

Specifications below reflect NVIDIA's public roadmap disclosures and will be updated as final details are confirmed.

SpecificationNVIDIA Rubin
PlatformVera Rubin — Rubin GPU + Vera CPU (custom Arm, ~88 cores)
MemoryHBM4 (next-gen high-bandwidth memory)
Rack SystemVera Rubin NVL144 — 144 GPU dies per rack
GPU InterconnectNVLink 6 — ~3.6 TB/s per GPU
NetworkingConnectX-9 (1.6T) / Spectrum-X & Quantum-X next-gen
CoolingDirect-to-chip liquid cooling (rack power rising sharply)
SuccessorRubin Ultra NVL576 (4-die packages, ~600 kW racks)
AvailabilityExpected 2H 2026 → 2027 (Rubin Ultra)

Plan for Rubin Now_

Power: design for 800V DC

Rubin-class racks push power well beyond today's 120 kW. New builds should plan 800V DC distribution and busway capacity ahead of delivery.

Cooling: liquid headroom

CDU capacity and facility water loops should be sized with headroom — Rubin Ultra trends toward ~600 kW racks. Under-provisioned cooling becomes a hard ceiling.

Fabric: 1.6T networking

Back-end fabric moves to ConnectX-9 / next-gen Quantum-X & Spectrum-X. Cable plant and tray routing should anticipate higher-radix, higher-bandwidth topologies.

Ready to Deploy Your GPU Infrastructure?_

Tell us about your project. Book a call and we’ll discuss scope, timeline, and the best approach for your deployment.

Book a Call