DRACO 128× H200 Modular Data Center

Overview

The DRACO 128× H200 Modular Data Center is a turnkey, modular AI data centre — 128× NVIDIA H200 SXM5 (18,048 GB HBM3e) across coupled containerized modules, pre-integrated with liquid cooling, UPS, power distribution, physical security and DCIM monitoring on a spine-leaf InfiniBand fabric. It stands up a full GPU data centre on a slab where there is no built facility — for sovereign programmes, neoclouds and large enterprises — and runs on-premises, in INR, on a GST invoice.

Engineered to deploy data-centre-scale AI without constructing a building, it couples self-contained modules into one validated system — GPU servers, fabric, liquid cooling, power and monitoring — delivered, integrated and burned-in as a single engagement with one warranty and one support contract.

Key highlights

• 128× H200 · 18,048 GB HBM3e — data-centre-scale AI capacity for training and inference, deployed on a slab.
• Modular, self-contained — coupled containerized modules with GPU servers, fabric, liquid cooling, UPS, PDU, security and monitoring.
• No building required — deploy where there is power and connectivity; expand by adding modules.
• 64× Intel Xeon 6 + 16 TB DDR5 ECC — host compute and memory matched to 128 GPUs.
• 2 PB NVMe NVMe + Spine-leaf InfiniBand NDR — large parallel storage and a non-blocking fabric for distributed training.
• DCIM remote monitoring + BMC/IPMI — full remote management of power, cooling, environment and compute.
• On-prem data sovereignty — data and models stay on site; DPDP-friendly, air-gappable.
• Make-in-India OEM — predictable INR pricing, GST tax invoice (HSN 8471), pan-India onsite support, GeM-procurable.

AI workload fit (what it actually runs — honestly)

• Training & inference at scale: distributed training and high-throughput serving across the 128 H200 GPUs over the InfiniBand fabric.
• Large-model serving: serve 70B–180B-class models, or host many large models concurrently.
• Vision, multimodal & agentic: organisation-wide production AI co-located with operations.
• Engineering note: this is a data centre, not a single system — modular construction lets you stand up 128 GPUs of liquid-cooled, battery-backed capacity on a slab in months, not years. We validate real scaling on your workload rather than quoting a peak number.

AI workload positioning

This sits at the deploy-a-data-centre stage: data-centre-scale AI without constructing a building. With 18,048 GB HBM3e across 128 GPUs plus its own power and liquid cooling, it is sized to sustain real training and serving at scale — the owned, sovereign alternative to leasing colocation or cloud.

Industry use cases

• Government & sovereign AI — a deployable national AI data centre on GeM-procurable infrastructure.
• Neocloud / providers — rapidly-deployed regional GPU data-centre capacity.
• Large enterprise & conglomerates — private AI data centres on owned land.
• Energy & resources — data-centre-scale AI at remote sites.
• Telecom — regional AI data centres at the network edge.
• Research & higher-ed — an institutional AI data centre without new construction.

Performance — and how to be sure

We don’t publish inflated peak numbers. The honest picture: 18,048 GB HBM3e across 128 H200 GPUs on a non-blocking fabric is sized for data-centre-scale training and serving. Want certainty? Request a free scaling benchmark of your model and dataset on a representative configuration before you commit; we’ll send back real tokens/sec, scaling efficiency and the full power and cooling envelope.

Series & upgrade path

• DRACO (flagship micro/modular-DC tier) — this, the largest in the line.
• Capacity ladder: pod → containerized → modular data center; expand by adding modules.
• When to step up: for fixed-site mega-scale, see RDP Rack-Scale AI Systems, SuperClusters and Supercomputers — talk to an architect about the build.

On-prem vs cloud — the TCO case

A modular data centre removes both construction time/cost and recurring cloud or colocation fees: you own the compute, power and cooling as one deployable asset, with no egress and full data residency. RDP pricing is fixed in INR with a GST input-credit-eligible invoice — ask for a multi-year TCO comparison including facility and construction savings.

Software & day-one readiness

Ships pre-integrated and validated: NVIDIA driver, CUDA, cuDNN, NCCL, the InfiniBand stack, Docker and NVIDIA Container Toolkit, with Kubernetes/Slurm, PyTorch and vLLM / Triton / TensorRT-LLM on Ubuntu LTS, plus DCIM monitoring. Optional managed operations and an MLOps platform.

Power, cooling & rack integration

A modular multi-container data center with liquid (CDU) cooling, integrated UPS, PDU and fire suppression — it provides its own thermal and power envelope; you supply site power and connectivity. (Exact power draw, BTU, battery runtime, slab and footprint figures confirmed in the design package.) Full DCIM and out-of-band management.

Deployment, warranty & support

• Made to order, integrated and burned-in in India; project timeline confirmed at quote.
• Delivered as one system: modules, GPU servers, fabric, liquid cooling, UPS/PDU, security, monitoring, and the full AI software stack.
• Onsite warranty + AMC with pan-India coverage, data-centre SLAs and a dedicated escalation path (exact terms confirmed at quote).

Why RDP

14 years of Make-in-India infrastructure and 300,000+ devices shipped. Indian OEM, INR pricing, GST tax invoice (HSN 8471), pan-India onsite engineers, GeM availability, and DPDP / sovereign-AI-ready deployment.

Buy with confidence

This is a turnkey modular AI data centre, made to order — talk to an RDP solution architect, size the modules, cooling and power for your site, get a multi-year TCO, and run a scaling benchmark before you commit. Request a quote to begin.

Specifications