Frontier Hardware
Racks of directly liquid-cooled NVIDIA B300 GPUs deliver frontier-scale training and inference at the highest density and efficiency available today.
The first rapidly deployable data center to respond to the AI energy crisis.
- 1.5MW
- Power Capacity
- 12
- Liquid-cooled Racks
- 720
- B300 GPUs
A rapidly deployable micro-data center.
Built in harmony with nature.
Secure Facility
A Faraday-caged enclosure shields every rack from interference and eavesdropping, while hardened steel entry, biometric access, and 24/7 surveillance lock your data down.
Extreme Quiet
Heat is spread across multiple loops and rejected through oversized dry coolers running well below peak speed, holding roughly 55 dB(A), about 75% quieter than conventional cooling.
N+1 Redundancy
Power, cooling, and networking all run at N+1 redundancy across the deployment, so any single component can fail and be serviced live without taking the site offline.
InfiniBand Networking
NVIDIA Quantum-X800 InfiniBand links nodes at 800 Gb/s per port over ConnectX-8 SuperNICs with SHARP v4 in-network compute, scaling training and inference near-linearly beyond Ethernet.
Compute that grows with you.
Every ZODE One site starts as a single rack and scales on demand. Drop in modules as your workloads grow, draw exactly the power you need, and run it all on renewable energy, without rebuilding from scratch.
Use existing energy
ZODE One taps the power you already have. Run on grid, solar, or on-site generation and scale draw to match your load, with no new substation required.
Deploy fast
Pre-integrated modules arrive ready to run. Site, connect, and bring compute online in days instead of the months a traditional build demands.
Respect the planet
Liquid cooling and renewable-first power keep efficiency high and emissions low, so every rack you add stays light on the land around it.
Built on Frontier-Class GPUs
Compute
720 NVIDIA HGX B300 GPUs
720 NVIDIA HGX B300 GPUs span 11 racks, with 288 GB HBM3e per GPU and NVLink 5 delivering full all-to-all bandwidth within each node.
Processors
Dual Intel Xeon 6 CPUs
Per server, dual Granite Rapids processors and up to 4 TB of DDR5 give the host headroom to drive all eight GPUs at full utilization.
Cooling
Closed-loop Liquid Cooling
A sealed, closed-loop liquid cooling system pulls heat directly off the GPUs and CPUs, sustaining peak clocks under continuous load while running quieter and denser than air.
Build
AI-Focused Design
Engineered from the ground up for AI inference, training, and other specialized workloads, with fast InfiniBand interconnect tying nodes together for low-latency, high-bandwidth scale-out.
Supercharge your ZODE.
ZODES can link to create more powerful sites.
1.5 MW
A single ZODE One. One module, sited and online in weeks, delivering 1.5 MW of dispatchable compute wherever the power already is.
9 MW
Link several units into a cluster. Shared power and networking turn a group of modules into 9 MW of coordinated capacity on a single site.
42 MW
Scale to a full ring. Dozens of ZODE One units combine into a 42 MW campus, all running on renewable-first power and liquid cooling.
ZODE One Specs
Dimensions
- Footprint
- 50 x 24 ft
- Square Footage
- 1,200 sq ft
- Height
- 24 ft
- Interior Volume
- ~28,800 cu ft
- Weight
- ~38,000 lbs
- Form Factor
- ISO-modular
50 ft24 ft24 ftCompute & Networking
- GPUs
- 720 NVIDIA B300
- Racks
- 12 liquid-cooled
- Rack Density
- 125 kW
- Fabric
- 400G InfiniBand
Power
- Capacity
- 1.5 MW dispatchable
- Source
- Grid + solar + storage
- PUE1
- < 1.1
- Backup
- On-site battery (BESS)
Cooling
- Method
- Direct-to-chip liquid
- Heat Rejection
- Closed-loop dry cooler
- Water Use (WUE)
- ~0 L/kWh
Deployment
- Lead Time
- 12 weeks
- Assembly
- Site-assembled
- Mobility
- Relocatable
1 Power usage effectiveness measured at full IT load under nominal ambient conditions. Specifications are representative and may vary by site configuration.
