DRACO ORv3 140 kW NVL72-Class AI Rack

Overview

The rack an NVL72-class system actually needs. A GB300 NVL72 or a 72-GPU rack-scale AI system draws roughly 135–142 kW sustained — well past what a 19-inch EIA rack on legacy AC distribution can feed or cool. This is the 21-inch OCP Open Rack v3 frame, 48 V DC busbar and direct-to-chip liquid loop that a rack-scale AI system is designed to drop into.

Key highlights

• 21″ OCP Open Rack v3 (ORv3) — the open standard NVL72-class racks are built on.
• 140 kW rack capacity — sized for a full rack-scale AI system, not retrofitted to one.
• 48 V DC busbar with blind-mate connectors — no per-server power cords, no stranded power.
• 4× 30 kW power shelves, N+1 — the ORv3 shelf architecture, AC→48V DC rectification.
• 480 V 3-phase input {VERIFY per site} with 2N upstream feed options.
• Direct-to-chip liquid ready — manifolds, blind-mate quick-disconnects, in-row CDU.
• Integrated, populated and commissioned in India by RDP.

AI workload fit

• Housing a GB300 NVL72 or equivalent rack-scale AI system.
• Housing a 72× MI455X or other 70-plus-GPU open-fabric rack.
• LLM training and inference at rack scale.
• Sovereign AI facility builds on open, non-proprietary rack standards.
• HPC & AI datacenters moving from 19″ EIA to OCP.
• AI-factory pods where every rack is identical and serviceable.

AI workload positioning

Most Indian datacenters are 19-inch EIA at 10–20 kW per rack. A rack-scale AI system needs roughly seven times that in a different physical standard, and the mismatch is where AI projects stall — the compute arrives and the facility cannot take it. Open Rack v3 (ORv3) is the Open Compute Project’s high-density rack standard: a 21-inch rack aperture instead of the legacy 19-inch EIA, a 48 V DC busbar running the full height of the rack, and blind-mate power connectors so servers slide in without individual power cords. Capacity is measured in OU (OpenU, 48 mm) rather than RU. The 48 V DC architecture raises usable power density substantially over legacy per-server AC distribution, which is what makes 100 kW-plus racks practical at all. RDP sells this as the matched pair to the compute, and will tell you plainly if your building is not ready for it. A facility survey is mandatory before order.

Industry use cases

• Neocloud: standardised, repeatable ORv3 rack units across a growing fleet.
• Public sector & sovereign: national AI facilities built on open standards, not one vendor’s rack.
• Enterprise & GCCs: the first rack-scale AI deployment in an existing datacenter.
• Research & education: housing a national or institutional rack-scale system.
• Telecom: AI capacity inside existing central-office and edge facilities.

Performance & how to be sure

The figure that governs everything here is kW per rack: 140 kW is roughly 7× a conventional 20 kW enterprise rack, and it is why liquid cooling and a 48 V busbar stop being optional. Rather than quote a tokens/sec figure that will not match your workload, RDP offers a “benchmark your model” session: bring the model, precision and context length you actually intend to run, and we will validate it on this exact configuration before you commit.

Series & upgrade path

Below: the DRACO 48U 100 kW 19″ EIA liquid-ready rack for dense-but-not-rack-scale builds. Above: the DRACO ORv3 480 kW AI Factory Rack. Pairs with RDP in-row CDUs and the DRACO 250 kW Power & UPS Block, and with Rack Integration & Deployment services.

On-prem vs cloud (TCO)

For sustained daily AI work this configuration removes per-GPU-hour billing, queueing for scarce instances, and egress charges on your own data — and keeps everything on-premises for DPDP and data-residency obligations. Cloud still wins for burst capacity and one-off very large training runs; on-prem wins on sustained utilisation, control and predictable INR capital cost. We model the crossover with your actual usage rather than assert it.

Software & day-one readiness

Rack-level monitoring covers busbar current, per-shelf load, coolant inlet/outlet temperature and leak detection, exported to your DCIM or to RDP’s managed monitoring. Power-shelf and CDU telemetry is available over Redfish/OpenBMC-compatible interfaces.

Power, thermal & acoustics

480 V 3-phase input to 4× 30 kW ORv3 rectifier shelves (N+1) feeding a 48 V DC busbar. Direct-to-chip liquid loop with rack manifold and blind-mate quick-disconnects; residual air load handled by rear-door HX or room cooling. Floor loading, water quality, flow rate and upstream switchgear are all surveyed before order — this is a facility project. Exact wattage, BTU and dB(A) figures come from RDP bench measurement rather than estimates — ask for the site-readiness sheet with your quote.

Deployment, warranty & support

Built to order by RDP Technologies. Supplied with GST invoice (HSN 8471), pan-India onsite support, and availability through GeM for public-sector procurement. Built to order — typically 10–14 weeks including facility survey and integration.

Why RDP

RDP Technologies is a Make-in-India OEM with 14+ years and 300,000+ units shipped, supplying AI infrastructure from desk-side systems to rack-scale AI factories — with predictable INR pricing, GST invoicing and pan-India onsite service.

Buy with confidence

Use Request a Quote to reach an RDP solution architect for sizing, a benchmark-your-model session, financing options and a delivery plan. No obligation.

Specifications