The Stack

Engineered from the fuel to the GPU

Most data centers were designed for web servers, then retrofitted for AI. The Data45 container is built the other way around — every system optimized for dense, sustained, liquid-cooled compute that can run at your site, on your power, under your control.

Energy Path

Power in, intelligence out

Every electron is accounted for. Follow energy as it moves from grid or behind-the-meter gas, through the liquid cooling loop and into 256 GPUs, then back out as model training and inference throughput.

  1. 1 MW
    Power In

    Grid feed or behind-the-meter natural gas, with a 48–72 hr LNG buffer.

  2. ΔT < 8°C
    Liquid Cooling

    Coolant distribution units pump dielectric fluid to every cold plate.

  3. 256 GPU
    Compute

    256 H200-class GPUs across 4 sealed hot-aisle racks, fully liquid-cooled.

  4. Fiber / Air-gap
    Fabric Out

    Connect over fiber and MPLS, or sever the link for a fully air-gapped run.

Core Engineering

Six systems, one integrated platform

Behind-the-Meter Power

Run on your existing grid supply or on behind-the-meter natural gas where the grid falls short — turning stranded energy into AI compute without waiting in an interconnection queue.

48–72 hr LNG Buffer

An onboard LNG buffer keeps the container running continuously through grid interruptions or fuel-supply gaps, so sustained training and inference never miss a beat.

Direct Liquid-to-Chip

A sealed liquid loop feeds cold plates on every GPU, removing heat at the source. That keeps a full 1 MW of dense compute stable inside a single 40-ft container.

Air-Gap & HSM Keys

The container supports a fully disconnected air-gap mode with no external connectivity. Customer-held HSM keys mean only your team can ever access the workloads.

Fiber, MPLS or Air-Gap

Connect over fiber and MPLS for hybrid operation, or sever the link entirely. The networking layer adapts to your security posture, not the other way around.

Fully Managed Ops

Data45 monitors and maintains the hardware layer — scheduled maintenance, hardware alerts, and GPU lifecycle management — backed by a written uptime SLA.

1 MW
Per container
256
H200-class GPUs
12 wks
Site-ready to live