AI EDGE
CUBE
Focus on AI. Leave the infrastructure to us. Power, cooling, IT, fire safety and control in a single 40-foot module — factory-built, factory-tested, connected on site.

§01Key figures
- IT load
- 300kW
- HGX B300 × 18 servers
- 144GPU
- Per GPU rack
- 135kW
- CDU capacity
- 600kW
- Liquid : air
- 9:1
- Contract → service
- 6mo*
* Six months assumes confirmed specs, secured power, permits and site readiness · Images are conceptual · Reference configuration as of Sept 2026
§02AI got fast. Infrastructure didn't.
| Driver | What changed |
|---|---|
| AI moves on-site | Physical AI, real-time inference and sovereign data processing |
| Power & land limits | New power and sites are hard to secure; legacy rooms fall short |
| Dense GPUs | Rising power and heat per server demand liquid cooling |
| Timing | Build when the GPUs arrive, not years ahead |
§03Six vendors. Or one.
Conventional procurement
- C-01Designer
- C-02Electrical
- C-03Mechanical
- C-04Fire
- C-05IT vendor
- C-06Controls SI
6 contracts · interfaces, schedule and risk sit with the owner
AI EDGE CUBE
- C-01DEUS — design, build, test, install, handover
1 contract · 1 window · 1 accountable party
§04Five systems, one enclosure
- S-01POWER
Switchgear · UPS
UPS 400 kW IT / 300 kW cooling, Li-ion 5-min backup, single 3-phase 380 V feed
- S-02COOLING
DLC + in-row air
Cold plate 90% · air 10% (N+1), CDU 600 kW, pumps N+1
- S-03IT
GPU racks 135 kW × 2
NVIDIA HGX B300 × 18 servers · 144 GPU, infra racks 15 kW × 2
- S-04SAFETY
Compartments · early detection
Battery fire compartment, separation zone, gas suppression, fire walls
- S-05CONTROL
DCIM
Digital twin thermal & airflow, CCTV and access control
§0524–36 months → 6
| M0 | Contract & spec | Rack density and availability tier fixed; long-lead items ordered |
| M1–M3 | Factory build + site prep | Enclosure, power and cooling assembled in the factory; utilities brought to site |
| M4–M5 | FAT · ship · deliver | Integrated load and cooling test, shipment, placement |
| M6 | Service start | Hook-up, SAT/IST, DCIM integration, handover |
§06Cooling
135 kW per rack. Cold plates carry 90% of the heat; in-row units handle the rest. A dedicated 300 kW cooling UPS keeps pumps and chillers running through outages.
| Method | DLC 90% · in-row 10% |
| CDU | 600 kW · pumps N+1 |
| Cooling UPS | 300 kW |
| Piping | HDPE / PP-H |
§07Fire containment
Batteries live in their own fire compartment. Six layers of passive protection keep a fire where it starts; end doors allow direct external suppression.
- P-01Thermo Coat — intumescent shell coating
- P-02Internal fire protection panels
- P-03Rack vent fire barriers
- P-04Cable & busbar fire barrier sheets
- P-05BULLSTOP penetration blocks
- P-06Fire sealing gaskets on hatches and doors
§08DCIM — one control plane
MODULE-01 ......................... ONLINE IT LOAD 286.4 kW ████████████████░░ PUE 1.12 CDU SUPPLY 32.1 °C OK RACK GPU-A 131.8 kW ████████████████▉░ RACK GPU-B 129.5 kW ████████████████▋░ BATTERY RM 24.0 °C OK · FIRE: CLEAR VIEW: THERMAL | MODEL | AIRFLOW | SIMULATE
- CONTROL
- Sensing · alarms · incident response
- OPERATE
- Assets · configuration · work orders
- INSIGHT
- AIOps · capacity · impact analysis
- SECURITY
- CCTV · access control · audit · RBAC
§09Deployments
ManufacturingReal-time AI next to the line
Vision inspection · predictive maintenance · robotics
Talk to us about this →
Public & researchSovereign research compute
AI research · shared GPUs · internal AI services
Talk to us about this →
Finance & enterprisePrivate AI and DR sites
In-house AI · disaster recovery · branch processing
Talk to us about this →
Defense & remoteDeploy fast, relocate later
Video & sensor analytics · edge inference · forward bases
Talk to us about this →§10Reference configuration
| Capacity | ≈ 300 kW IT load |
| Redundancy | N baseline / N+1 · 2N by design |
| GPU rack | 135 kW × 2 (ORv3, 1400 A busbar) |
| Infra rack | 15 kW × 2 |
| Compute | NVIDIA HGX B300 · 18 servers / 144 GPU |
| UPS | 400 kW (IT) / 300 kW (cooling) |
| Battery | Li-ion · 5 min · dedicated compartment |
| Generator | Optional |
| Cooling | DLC 90% (CDU 600 kW) / in-row 10% (× 4, N+1) |
| Power feed | 3-phase 380 V 60 Hz, single line |
| Module | Wide 40-ft-class enclosure · outdoor chiller |
| Schedule | Contract → service ≈ 6 months* |
- A-01
Truly modular
Relocate when the site changes. Lease and subscription ready.
- A-02
One power feed
A single 3-phase 380 V line feeds everything from servers to chiller.
- A-03
Cooling never stops
A dedicated 300 kW cooling UPS keeps pumps and chillers running.
- A-04
Fire stays contained
Battery compartment plus six layers of passive protection.
- A-05
Your GPUs, your cloud
Customer-chosen GPUs and stack. Air-gapped operation supported.
- A-06
One contract
Design, build, test, install and handover — one accountable partner.
Reference as of Sept 2026; subject to manufacturer changes and project conditions.
§11Built by a full-lifecycle team
- 10+
- Data centers designed & built
- 15+
- Feasibility & basic designs
- 21+
- Sites reviewed
§12Proposal and quote in about two weeks
- 01Share requirements
Location, available power, servers, target date
- 02Reference model & options
Schedule and quote in about two weeks
- 03Lock spec & scope
Supply and site-work scope agreed
- 04Order & start
Long-lead equipment ordered; build starts at M0