SuperPod Ultra
One pod. Sixteen Vera Rubin racks.
The next generation of HYPERCUBE: a single-pod reference architecture targeting 1,152 NVIDIA Rubin GPUs, with compute, fabric, storage, liquid cooling and protected power designed as one integrated system.

Styling concept. Configuration not finalised.
StatusReference architecture. Design targets. Not yet built. At engineering review, before product basis freeze. Planned for Q2 next year, after certification.
Configuration VR72-1152. Final equipment, ratings, availability and delivery remain subject to qualification and written confirmation.
One object. Not a yard of boxes.
A continuous white sleeve that is never punched through, a graphite core that carries every industrial interface, and one emerald line where the two meet.
One language, at every size.
The same width, the same materials and the same detailing, from a first eight-bay step to a protected campus.
Inside, the same discipline.
The compute row, the support positions and the hydraulic plant inside one sleeve, served by one aisle and one overhead manifold run.
Made in a factory. Finished near the customer.
The sleeve ships as cassettes and the structure as a frame, so every module is built, inspected and tested the same way.
At home in any setting.
The form does not change. The corrosion class, fixings and site treatment are specified for the destination.
Styling concept from the HYPERCUBE SuperPod Ultra Design Collection, Edition 2. Configuration not finalised; renders do not depict a supplied configuration.
Cooling, in the same language.
Three open-air thermal packages beside the hall, dressed like the hall: a sleeved fan deck or a graphite louvre fascia.
After dark.
The emerald line runs along the base of every module and traces the whole pod from a distance.
Styling concept from the HYPERCUBE SuperPod Ultra Design Collection, Edition 2. Configuration not finalised.
Designed for the next generation.
Reference values for one VR72-1152 pod.
Compute architecture
- Rack target
- 16 × NVIDIA Vera Rubin NVL72
- Processor target
- 1,152 Rubin GPUs and 576 Vera CPUs
- Layout
- Sixteen racks in one row within a single widebody compute module
Power design
- IT envelope
- 5.98 MW: 5.28 MW compute plus 0.70 MW support IT
- Rack reservation
- 330 kW per rack at the full facilities design reservation
- Site planning
- 10.5 MW connection allowance in the reference design; site input is higher than IT load
Integrated infrastructure
- Fabric
- Eight-rail 800G scale-out design joining sixteen separate NVLink domains
- Cooling
- Direct liquid cooling with coolant distribution, buffer and external heat rejection
- Protected power
- UPS, distribution and battery ride-through within the pod scope
Factory and site
- Manufacture
- Planned for South Korea, with integrated factory acceptance before shipment
- Transport
- Six oversize enclosures, 4.5 m wide, plus three open-air chillers
- Site scope
- Power at the intake, foundations, roads, drainage, permits and fire services
All values describe the HYPERCUBE SuperPod Ultra reference architecture and are design targets, not measured pod performance or a supplier quotation. Final configuration and operating ratings depend on equipment selection, destination requirements, engineering qualification, factory acceptance and site acceptance. The image is a concept visualization.
Nine packages.
One product.
Six enclosed modules and three open-air chillers form the planned pod. The installed package count is not a shipping-container count.
Six enclosed modules
- C1 · Compute
- Sixteen Vera Rubin NVL72 racks
- S1 · Fabric and storage
- Cluster networking, performance storage and context tier
- H1 · Hydraulics
- Coolant distribution, pumps and thermal buffer
- U1 · Protected power
- UPS banks and distribution
- B1 · Battery
- Battery ride-through provision
- M1 · Intake
- Transformers and incoming power interface
Three heat-rejection packages
- T1 to T3 · Chillers
- Three open-air units, with any two intended to carry the design cooling duty
- Transport
- The nine installed packages may require more than nine shipping consignments
- Power continuity
- No generator is included in the pod. The cooled battery ride-through target remains to be proven in acceptance testing
From architecture
to accepted pod.
Engineering review is the current stage. No SuperPod Ultra unit has yet been built.
- Freeze the product basis and qualify supplier interfaces
- Prove service access with full-scale mock-ups and release fabrication drawings
- Accept each module and test the complete pod together under load and defined fault conditions
- Approve transport, complete site acceptance and issue destination-specific operating ratings
Named technology platforms are reference selections for qualification. This architecture is not a statement of approval or endorsement by NVIDIA, Supermicro or any other supplier, and does not constitute a performance guarantee.
Plan for SuperPod Ultra.
Share your intended location, available power and workload requirements to discuss whether the reference architecture fits your next-generation AI programme.