Physical AI workload
Video, sensor fusion, simulation, inference and model update have different infrastructure needs.
We begin with workloads, data ownership and regional operations rather than selling a box or a GPU list.

Video, sensor fusion, simulation, inference and model update have different infrastructure needs.
Decide what remains at the device, regional node and core.
Protect and govern how field data becomes learning assets.
Connect power, cooling, network, buildings and operations.
Keep confidential data inside a controlled computing environment.
Support local services and continue operating during disasters.
Phase power and capacity in line with anchor workloads and demand.
Recognize and act immediately near cameras, drones, robots and facilities.
Share selected context, update regional models and coordinate sites.
Retain, analyze and govern multi-region data and models.
Factory-integrated power, cooling and data-hall modules can be tested before site delivery. A campus can expand in 1 MW pod increments toward a 20 MW concept.

5 racks / 27 kW reference base design.
6 racks / 38 kW reference base design.
12 racks / 80 kW reference base design.
14 racks / 94 kW reference base design.
Figures are reference configurations, not guarantees. Final design depends on equipment, redundancy, climate and site conditions.
Utility intake, switchgear, transformers, UPS, busway and BESS are designed as one topology.
Direct liquid cooling, CDU and facility-water loops are selected for rack density.
Reserve space, power paths and cooling capacity for technology change.
High-density rack concepts require structural, hydraulic and redundancy verification.

Closed networks, domestic infrastructure, access control, audit logs and model governance let organizations use AI without surrendering sensitive assets to uncontrolled external services.
Distributed nodes and local continuity planning.
Shared infrastructure for local companies and public services.
Coordinate renewable energy, storage, grid constraints and AI demand.
Plan land, water, noise, employment and public value together.
Confirm utility, generation, storage and expansion paths.
Add capacity in controlled increments.
Design network, cooling and operations for target workloads.
Factory integration can shorten site construction and commissioning.
Start from identifiable demand rather than speculative capacity.
Workload, data architecture, Japan integration and business design.
Power systems, cooling, modular infrastructure and international delivery experience.

Partnership development, customer discovery, regional discussions and modular design studies proceed in stages. Capacity, location, schedule and commercial terms are not presented as committed until contracts and approvals are complete.
Sovereign AI means using AI while keeping data, models and IP inside Japan. Through our strategic partnership with GEC (Taiwan) for modular data centers and our vertically integrated stack from edge to data infrastructure, we take Sovereign AI from concept to implementation. We seek adoption partners (manufacturing, finance, healthcare, municipalities), build-out partners (land, power, construction), and sales partners. And we will be user No. 1 — migrating our own AI development and operations onto the same infrastructure.
The service starts with workload and data architecture, then selects computing and infrastructure.
Yes. All-in-one modular nodes provide smaller reference starting points.
Yes. Cooling topology is selected according to rack density and equipment.
An AI environment designed to keep data, models, access and governance under required domestic or organizational control.
Yes. Regional use, resilience, energy and community value are considered together.
No. They are planning references; final values require engineering and contract confirmation.
Choose the inquiry path for enterprise, municipality, investor or development partner.