Regional Network
AI Factory Sites
Drag to rotate · hover for detail
2 MW initial
Target expansion to 10 MW. Planned deployment includes NVIDIA B300 server clusters and Pro 6000 inference card clusters.
0.75 MW initial
Target expansion to 5.5 MW. Planned deployment focuses on NVIDIA Pro 6000 inference card clusters.
2.4 MW initial
Target expansion to 30 MW. Planned deployment includes NVIDIA B300 server clusters and Pro 6000 inference card clusters.
Systems built for scale.
01 · BUILD TIME
180DAYS
From design to complete construction of the AI Factory.
02 · OPERATIONAL
30DAYS
From equipment installation to a fully operational facility.
03 · DEPLOYMENT MODEL
PUE
<1.2
Power density
Up to 1.5MW / 2000 ft²
GPU per node
>32
TWO CONTAINERIZED PATHS
Built around
cooling architecture.
01
Liquid-cooled B300
A single 1.5 MW deployment group — compute, CDU cooling, power, and UPS packaged together for high-density liquid-cooled servers.
02
Air-cooled B300
A modular 1 MW scheme — two compute containers with shared power distribution and UPS for flexible capacity planning.
MODULAR CONTAINER ROOMS
High-density compute, faster to operate.
NVIDIA B300 SERVERS
Prefabricated modular container-room solutions for high-density NVIDIA B300 server deployments, co-developed with global partners.
CONTAINER FORMAT
1 modular container
POWER CAPACITY
1.5 MW
GPU DENSITY
512 GPUs
TARGET WORKLOAD
Training & inference
—Designed around NVIDIA B300 server clusters.
—Containerized compute increases token output per unit area.
—Supports liquid cooling and advanced thermal design.
NVIDIA PRO 6000 INFERENCE
A two-container modular design that creates dense inference capacity with manageable power and thermal requirements.
CONTAINER FORMAT
2 modular containers
POWER CAPACITY
1 MW
GPU DENSITY
960 GPUs
TARGET WORKLOAD
High-throughput inference
—Designed around NVIDIA Pro 6000 inference card clusters.
—Two-container config supports high GPU count and modular expansion.
—Optimized for inference density and token output per unit area.