Data
Centers.
Place your GPUs in a ready data center. From a single node to full clusters.
We source high-density space, install DGX and HGX systems, plan data center readiness and support your hardware on site with Smart Hands.
- High density
- 200 kW – 1 GW
- Starting from
- 1 node
- GPU platforms
- H200 · B300 · GB300 · Vera Rubin
- Smart Hands
- Bangkok · Chonburi · Rayong
High-Density GPU Colocation
Start with one node. Scale up to a gigawatt.
We source space, power and cooling for high-density NVIDIA GPU systems. Place a single GPU server, or reserve high-density capacity from 200 kW up to 1 GW.
Power per node
Match space and cooling to the power each platform draws.
On smaller screens, scroll the table sideways to see every column.
| System | GPUs | Footprint | kW per node | kW per rack | Cooling | Sources |
|---|---|---|---|---|---|---|
| DGX H200 | 8 × H200 | 8U per node | 10.2 kW | ≈ 41 kW · 4 nodes | Air | DGX H100/H200 User GuideNVIDIA H100/H200 RA · rack density |
| DGX B300 | 8 × B300 | 10U per node | 14.5 kW | ≈ 58 kW · 4 nodes | Air | DGX B300 User GuideNVIDIA B300 RA · rack density |
| GB300 NVL72 | 72 × Blackwell Ultra · 36 Grace | 1 rack · 18 compute trays | ≈ 7.9 kWPer compute tray · average | Up to 142 kW | Fully liquid | NVL72 AI Factory ERAGB300 NVL72 |
| Vera Rubin NVL72Preliminary | 72 × Rubin · 36 Vera | 1 rack · 18 compute trays | ≈ 10.0–12.2 kWPer compute tray · average | ≈ 180–220 kWEstimate | Liquid · 45 °C water | DGX Vera Rubin NVL72 (preliminary)SemiAnalysis · kW estimate |
Starting reference values from manufacturer documentation; confirm them with your OEM and data center. NVL72 per-tray values divide total rack power, including NVLink switches. NVIDIA has not published Vera Rubin NVL72 rack power; the 180–220 kW range is an industry estimate. OEM HGX servers vary by model.
Cluster size by power capacity
See how many nodes and racks each power envelope can hold.
On smaller screens, scroll the table sideways to see every column.
| Power | DGX H200 | DGX B300 | GB300 NVL72 | Vera Rubin NVL72 |
|---|---|---|---|---|
| 200 kW | 5 racks19 nodes · 152 GPUs | 4 racks13 nodes · 104 GPUs | 1 rack18 nodes · 72 GPUs | Below one rack |
| 1 MW | 25 racks98 nodes · 784 GPUs | 17 racks68 nodes · 544 GPUs | 7 racks126 nodes · 504 GPUs | 4 racks72 nodes · 288 GPUs |
| 10 MW | 245 racks980 nodes · 7,840 GPUs | 173 racks689 nodes · 5,512 GPUs | 70 racks1,260 nodes · 5,040 GPUs | 45 racks810 nodes · 3,240 GPUs |
| 100 MW | 2,451 racks9,803 nodes · ≈ 78,400 GPUs | 1,724 racks6,896 nodes · ≈ 55,200 GPUs | 704 racks≈ 12,700 nodes · ≈ 50,700 GPUs | 454 racks8,172 nodes · ≈ 32,700 GPUs |
| 1 GW | ≈ 24,500 racks≈ 98,000 nodes · ≈ 784,000 GPUs | ≈ 17,200 racks≈ 69,000 nodes · ≈ 552,000 GPUs | 7,042 racks≈ 127,000 nodes · ≈ 507,000 GPUs | 4,545 racks≈ 81,800 nodes · ≈ 327,000 GPUs |
Counts compute IT load only, using the power figures above. Network, storage, management and facility overhead such as cooling and UPS losses are excluded. H200 and B300 assume four nodes per rack. NVL72 counts each four-GPU compute tray as a node, with Vera Rubin sized at 220 kW per rack.
Data center tiers
Choose the standard that fits your workload and budget. We compare providers before we quote.
Premium
Higher design and operating standards, priced above the other tiers.
Best forCritical systems with strict compliance requirements
Standard
A balance of facility standards and cost.
Best forMainstream production workloads
Essential
A lower entry cost with room to grow in phases.
Best forBudget-controlled workloads that start small
Data center providers are named in your quotation.
Available power capacity
Check high-density power ready for installation. We refresh these figures regularly and confirm them in every quotation.
On smaller screens, scroll the table sideways to see every column.
| Tier | Location | Available | Max kW per rack | Cooling |
|---|---|---|---|---|
| Premium | Pending | Pending | Pending | Pending |
| Standard | Pending | Pending | Pending | Pending |
| Essential | Pending | Pending | Pending | Pending |
Last updated: Pending
How pricing works
CPU colocation has standard monthly pricing. GPU high density is quoted per project.
CPU Colocation
Flat monthly pricing for CPU servers, like conventional colocation.
| Package | Premium | Standard | Essential |
|---|---|---|---|
| 1U | Pending | Pending | Pending |
| 1/8 Rack | Pending | Pending | Pending |
| 1/4 Rack | Pending | Pending | Pending |
| Full Rack | Pending | Pending | Pending |
GPU High Density
Priced around power and each GPU architecture.
- Monthly Power
- Pay for power monthly. Rates depend on the data center tier and the power you use.
- One-time Customization
- Adapt the space to your GPU platform. Power, cooling and rack changes are a one-time charge.
- Contract Term
- High-density contracts start at three years. Term and renewal conditions are set in the quotation.
- Other Charges
- Other charges follow the agreed scope. Each is itemized in the quotation.
Rack & Stack · Commissioning
GPU clusters racked, cabled and ready to run.
We install DGX and HGX systems, cable within and between racks, configure rPDUs and verify every connection before handover.
DGX / HGX Installation
Mount GPU servers to the planned rack elevation. Weight, position and airflow follow the manufacturer’s requirements.
Rack Elevation · Rails · Airflow
rPDU & Power Cabling
Feed every system from both power sources. Configure rPDUs and route A/B power cords to the redundancy plan.
rPDU · A/B Feed · Phase Balance
In-rack Cabling
Connect GPU servers with MPO and LC fiber and RJ45 Cat6A. Every cable is dressed, labeled and recorded.
MPO · LC · Cat6A · Labeling
Inter-rack Cabling
Connect every rack to the OOB, in-band and GPU fabric networks. Cables follow the cluster cabling matrix.
OOB · In-band · GPU Fabric
Commissioning
Hand your platform team a verified cluster. Power, BMC access and every link are checked against the cabling matrix.
Power-on · BMC · Link Validation
CPU Server Installation
Install CPU servers in your medium and small data centers. The same cabling and documentation standards apply.
Rack & Stack · Cabling · Documentation
Our team has installed DGX and HGX systems alongside NVIDIA installation teams.Team experience
Data Center Readiness
Ready the facility before the GPUs arrive.
We advise data center operators and organizations planning NVIDIA GPU deployments, working with your team from power planning to installation day.
Data center operators
Prepare your halls for high-density GPUs. Assess power, cooling and rack layout against the platforms your customers bring.
Organizations planning GPUs
Choose a facility and plan installation around the GPUs you buy. Check site readiness before hardware is ordered.
Power & Cooling Assessment
Know the facility can host your chosen GPUs. We assess power per rack and air or liquid cooling options.
kW per Rack · Air · Liquid
Rack Layout & Floor Planning
Lay out racks for weight, cable paths and growth. We check floor space, floor loading and rack access.
Floor Loading · Aisles · Growth
Network & Cabling Design
Define every network path before a cable is pulled. We design cabling for GPU fabric, in-band, OOB and storage.
Cabling Matrix · Fabric · Storage
Installation Planning
Bring equipment on site on one shared plan. We sequence receiving, installation and acceptance with the data center and vendors.
Receiving · Sequencing · Acceptance
Smart Hands
On-site engineers who know GPUs, working to your SLA.
We operate and support GPU servers, CPU servers and networks in the data center for your team, across Bangkok, Chonburi, Rayong and the EEC.
GPU & CPU Server Support
Resolve hardware faults within your SLA. We diagnose issues, replace parts and coordinate vendor RMAs.
Diagnostics · Parts Replacement · RMA
Network Support
Bring failed links back into service. We check cables, transceivers and ports across GPU fabric, in-band and OOB.
Transceivers · Cabling · Ports
AI Factory Operations
Keep GPU clusters running with a team that has operated AI factories, from the rack to the fabric.
GPU Cluster · Fabric · Runbooks
Site Access Preparation
Start work as soon as we arrive. Access requests, spares and method statements are prepared to each data center’s process.
Access Request · Spares · MOP
- Coverage
- Bangkok · Chonburi · Rayong · EEC
- Neighboring countries on request; travel time applies.
- Hardware we support
- Dell · HPE · Supermicro · Lenovo
- And other vendors
- Team experience
- Our team has worked in several large data centers across Thailand and knows each site’s access process.
From the data center to your AI platform
Once your GPUs are racked, continue with cluster, network and platform design for model serving.
Tell us your GPUs and the power you need.
Share your GPU platform, node count and preferred location for a colocation, installation or Smart Hands proposal.
Request a quote