Power and connectivity
Confirm available service, energization timing and network connectivity. The GB300 reference uses an approximately 3 MW gross facility interface per cluster.
Alpha
Turnkey compute.
Deployed within 180 days.
A modular data center for AI, with power, cooling and rack infrastructure integrated before delivery.
From factory integration to site commissioningExplore the system
- 2.5 MW
- Reference IT peak per cluster
- 24
- Rack positions
- 1,152 GPUs
- GB300 NVL72 reference per cluster
Build in parallel.
Connect on site.
At the factory
Build and test the cluster.
At the site
Prepare foundations, power and connectivity.
Then deliver, connect and commission.
Factory integrated.
Faster to load.
- 01
Rack infrastructure, coordinated.
Rack positions and service access are planned as one layout, with space for equipment installation and ongoing maintenance.
- 02
Power distribution, integrated.
Overhead A/B busways bring distribution into the enclosure, aligning power connections with the rack layout before delivery.
- 03
Cooling interfaces, defined.
Supply and return headers bring liquid cooling to the rack row, with interfaces coordinated for the selected rack platform before site connection.
Faster to load.
Flexible to scale.
Add Alpha clusters as power and demand grow. Plan the first phase around your workload, then reserve land, electrical capacity and cooling connections for expansion.
2.5MWReference IT peak
1 cluster / 24 rack positions
Start with the site.
Design for the load.
Land and access
Allow space for the cluster, external cooling plant, delivery vehicles, lifting and maintenance. Reserve the land and interfaces needed for later phases.
Cooling and climate
The reference design uses liquid cooling and dry heat rejection. Hot conditions can require adiabatic assist using stored water; storage and replenishment follow the site climate.
Rack and operating needs
Set the rack platform, sustained load, redundancy and operating requirements. Final equipment interfaces and local approvals are part of project engineering.
Plan the first phase.
Prepare for what follows.
Share your target capacity, location and timing. We’ll review your site and capacity requirements, identify open questions and recommend who should join the engineering discussion.
Alpha
technical reference.
Review the GB300 reference below, or open the product overview for the system scope, site inputs and delivery sequence.
Compute
- GB300 NVL72 rack positions
- 16
- Network and support rack positions
- 8
- GPUs, reference configuration
- 1,152
- Grace CPUs, reference configuration
- 576
- Modeled rack power, GB300 EDPp
- ≈155 kW
Power
- Modeled electrical peak, IT
- 2.536 MW
- Reference gross facility interface
- ≈3.0 MW
- Busway provisioning per GB300 rack
- 192 kW
- Distribution basis
- Independent A/B paths; optional 2N UPS
- UPS autonomy, budgetary basis
- 15 minutes
Cooling
- Direct liquid cooling design capacity
- 2.100 MW
- Air side design capacity
- 0.280 MW
- Total thermal design basis
- 2.380 MW
- Heat rejection duty capacity
- 2.500 MW
- Facility water supply and return
- 41/51 °C
Water and footprint
- Normal operating mode
- Dry, no water
- Adiabatic transition reference
- ≈39 °C ambient
- Utility water connection
- Not required
- On site storage equivalent
- 100 m³
- IT module envelope
- 16.2 m × 6.8 m
Electrical peak, thermal design and heat rejection duty describe different reference conditions. Project engineering must reconcile them for the intended sustained load. Installed feeder capability is not simultaneous IT consumption.
View technical overview