Workload
Separate document search, video inference and training; check concurrent users, data volumes, GPU and memory demand, and operating hours.
WORKLOAD-FIRST AI INFRASTRUCTURE
Design the data center
Design the data center
around how AI will be used.
AI Infrastructure, Designed from the Workload Out.
Starting with Physical AI data from cameras, Edge AI, drones, robots and industrial equipment, Standard Link designs communications, power, cooling, GPUs and distributed AI data centers together.
Combined with GEC's modular power, cooling and EPC technology, this is a vision for infrastructure balancing closed-network operation, scalability and high-density AI processing.
- WORKLOAD-FIRST
- SOVEREIGN
- MODULAR
- DISTRIBUTED
Discuss a PoC or joint development
Discuss regional AI infrastructure
Discuss partnerships or investment
SOURCE IMAGE / GEC PROFILE P.29 / MODULAR DATA CENTER CONCEPT
Assess the AI workload before sizing the facility.
The 20-foot, 5-rack, 27kW configuration on our public site is a reference starting point. The definition of 27kW, GPU selection, cooling, installation and maintenance are confirmed for each project.
Power
Check site power availability, redundancy, UPS and the scope and definition of power ratings.
Cooling
Check expected heat, cooling method, outdoor conditions, noise and maintenance access together with the GPU configuration.
Phasing
PoC, small-scale introduction and staged expansion as usage grows.
If Edge AI or local processing alone meets the objective, the assessment may recommend not introducing an AIDC.
Discuss a small-AIDC preliminary assessmentThese are reference configurations for design and construction discussions, not facilities owned or operated by SLL. The parties responsible for design, supply, installation and operations are confirmed per project.
Training, adaptation of existing models, evaluation, multi-video inference and data management are different workloads. We also assess whether existing servers or cloud services are sufficient; this small configuration is not presented as capable of every training task.
- Who it is for
- Companies and research institutions planning AI processing or data management at factories and research sites.
- What we can discuss
- Workloads, site, power, cooling, communications and the division of maintenance and operational responsibility.
- Deliverables after agreement
- Workload summary, configuration comparisons, site conditions and responsibility matrix. These are not completed construction or operating records.
- Fit conditions
- We define configurations and support scope from the use case, site, power and cooling conditions. The definition of the 27kW power rating is confirmed individually.
- Next steps
- Consider existing facilities, staged-configuration evaluation and project-specific installation and operating design.
02 / WHY STANDARD LINK
Design from real AI demand, not from the enclosure.
Rather than starting with power, racks, cooling and connections, Standard Link works backwards from the Physical AI workload: what AI vision and Edge AI capture, where decisions are made and what is learned.
WORKLOADSee and sense
EDGE AIDecide and control
NETWORKClosed and distributed
POWERHigh density and redundancy
COOLINGAir and liquid cooling
AI CORETrain and distribute
Physical AI Workload
Design around real workloads from cameras, sensors, drones and industrial equipment.
Edge-to-Core Architecture
Design Device Edge, Regional AI Nodes and core AI data centers as one operating system.
PUBLIC-EVIDENCE Data Pipeline
Use IP to support seeing, detecting, integrating, deciding, recording and learning.
Infrastructure Integration
Integrate GEC power, cooling and modular EPC with Japanese applications, regulations and partner arrangements.
03 / THREE AI INFRASTRUCTURE SOLUTIONS
Different users need different AI infrastructure.
Match closed-network design, placement, power, cooling and expansion units to confidentiality, regional needs, investment returns and actual demand.
01 / PRIVATE AI INFRASTRUCTURE
Dedicated AI computing that keeps confidential data within the environment.
Manufacturing / research / healthcare / finance / critical infrastructure / AI developers
- Closed networks and dedicated GPUs
- Domestic installation and customer-specific data separation
- Staged expansion from PoC
- IP and research-data protection
02 / REGIONAL AI INFRASTRUCTURE
Distributed AI for regional challenges and continuity during disasters.
Local government / industrial parks / ports and airports / universities / regional power / agricultural regions
- Disaster-response AI and smart cities
- Regional processing and shared use
- Regional industry digital transformation
- Renewable energy and BESS integration
03 / AI INFRASTRUCTURE DEVELOPMENT
AI infrastructure with staged investment and capacity expansion.
International investors / developers / infrastructure funds / general contractors / energy companies
- Modular, staged investment
- Time to Revenue
- High density and liquid cooling
- BESS and Japanese-market integration
04 / EDGE-TO-CORE DISTRIBUTED AI
Decide at the edge, coordinate regionally and learn at the core.
Instead of continuously sending all raw data to a central location, select data at the edge and share necessary information in stages according to the application and contract.
Recognize on site and act promptly.
HOLONIC Eyes, cameras, sensors, drones, robots, factory and agricultural equipment, and HOLONIC Edge AI.
- Immediate recognition and anomaly detection
- Data selection and local storage
- Equipment control
- Autonomous operation during communications loss
Coordinate intelligence at the regional level.
An intermediate layer that integrates multiple sites locally for shared use with lower bandwidth demand and latency.
- Multi-site integration
- Regional data sharing
- Low-latency analysis and medium-term storage
- Regional model operations
Learn across regions and return improved intelligence.
A closed-network, sovereign core handles long-term storage, integrated analysis, training, distribution and audit.
- AI model training and simulation
- Long-term storage and integrated analysis
- Model distribution
- Data governance
- CAPTURECaptureCamera / Sensor / Drone
- UNDERSTANDUnderstandHOLONIC Edge AI
- SHAREShareRegional AI Node
- LEARNLearnHOLONIC DATA CENTER
- DEPLOYDistributeModel Update / OTA
- IMPROVEImproveImprove field accuracy
05 / MODULAR AI DATA CENTER
Start small and expand with demand.
Power, cooling and data-hall modules are manufactured and integration-tested in a factory, then delivered while on-site construction proceeds in parallel. Capacity is added as AI demand grows to limit initial investment.
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
Prefabricated AI data center solution
1. Power module
- Capacity: 625 kW–2 MW
- Galaxy UPS with lithium batteries
- Okken / Blokset panels
- Cooling: CHW / DX types
- Fire suppression system / smoke detection system
2. Data hall
- Multi-bay layout, up to 200 racks
- Busway power distribution
- Racks and metered power distribution
- EcoStruxure DCIM software
- Environmental monitoring and security
- Smoke detection system
3. Cooling technology
- Air cooling: below 40 kW per rack
- Direct liquid cooling (DLC): up to 140 kW
- Schneider CDU: 1000 kW
- Hybrid cooling combining DLC and air cooling
- Water-cooled or air-cooled chillers
The three illustrations show a power module, an internal rack row and an overall facility layout. The numbered list on the right groups power, data-hall and cooling functions. Schneider Electric branding remains part of the supplier image.
Site note: these are supplier reference configurations, not a guarantee for every project. Capacity definitions, applicable equipment and site conditions require confirmation.
GEC REFERENCE CONFIGURATION
Power, cooling and data halls arrive after integration testing.
- POWER MODULE
- 625kW - 2MW
- DATA HALL
- Up to approximately 200 racks
- COOLING
- Air / direct liquid / hybrid cooling
- SCALE
- Add 1MW pods in stages
- DESIGN TARGET
- Configuration targeting PUE around 1.2
Figures describe GEC reference configurations. Actual capacity, rack count, cooling and PUE depend on equipment, climate, redundancy, site and power conditions.
Configure power reception, transformation, UPS and storage for the project.
Integrate racks, distribution, monitoring and security.
Select air or direct liquid cooling for density and location.
Image supplied from PDF page 18. The specification summary on the right is adapted for the web; capacity and methods vary by project.
TIME TO OPERATION / CONDITIONAL TARGET
Target: operations in eight months.
This is a target for projects with site, power, permits, procurement and other prerequisites in place, where factory prefabrication and on-site construction run in parallel. Completion and start-up dates are not guaranteed.
20MW REFERENCE ARCHITECTURE
Scale 1MW pods to twenty as demand grows.
Staged-expansion concept using N+1 power and cooling blocks. Piping and equipment details are designed individually.
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
AIDC modular cluster layout
20 MW for an IT campus: twenty 1 MW pods, with an N+1 configuration. The source layout specifies 90% heat removal through liquid cooling, warm water at 30–35°C, and a target PUE of 1.2.
Legend
- Light blue: dry-cooler yard (warm water)
- Pale yellow: UPS and batteries (container)
- Pale violet: pump / heat-exchanger (HEX) skid (container)
- Gray: building / white space
Approximate site boundary / outdoor energy and cooling yard: four 5 MW clusters
Clusters 1, 2, 3 and 4 are each labeled 5 MW.
- Each cluster contains dry coolers (N+1), a pump / HEX skid (N+1), and UPS plus batteries (N+1).
Data center building: white space, CDU and pod layout
Pods P01–P20 are each labeled 1 MW, arranged in four rows of five: P01–P05, P06–P10, P11–P15 and P16–P20.
- Access road / truck route
Within the approximate site boundary, the four outdoor clusters occupy the upper zone; the building with twenty pods is below. The access road / truck route runs along the bottom.
Site note: this is a reference layout. The heat-removal ratio and PUE are design figures or targets, not verified operating results or project-wide guarantees.
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
Modular data center
- Flexibility
- Speed
- Predictability
The left side lists the three intended benefits and shows the overall facility. The right side illustrates cooling, power and rack modules. Schneider Electric marks remain in the supplier image.
FLEXIBILITYFlexibility
Start with the capacity needed and add as demand grows.
SPEEDSpeed
Run factory manufacturing and site works in parallel to aim for earlier operation.
PREDICTABILITYPredictability
Manage quality and schedule through integration-tested modules.
Key points from PDF page 30 adapted into three web cards.
Start with five racksand expand with AI demand.
Standard Link's AI data center approach is not limited to building a large facility all at once.
From all-in-one AI nodes in the tens-of-kW range at factories, universities, hospitals, local government and research sites, to dedicated enterprise or regional data centers of 625kW to 2MW and 20MW-class AI campuses, expansion is staged around purpose, power conditions and AI demand.
Small container-based edge AI data centers can serve as an entry point for enterprise and local-government AI infrastructure.
AI computing at the scale and location required.
Start small and expand in line with usage and revenue.
GEC REFERENCE / INTERNAL CUTAWAY
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
All-in-One Modular Data Center
Prefabricated IT infrastructure: UPS, cooling, racks and management software.
A manageable all-in-one solution assembled at the factory in a secure, weather-resistant structure.
- Enclosure formats: 20-foot and 40-foot ISO containers; 25-foot and 45-foot modules.
Four base configurations
- 20-foot ISO container: 5 racks / 27 kW.
- 40-foot ISO container: 12 racks / 80 kW.
- 25-foot non-ISO container: 6 racks / 38 kW.
- 45-foot non-ISO container: 14 racks / 94 kW.
Equipment labels in the cutaway
- Cable containment
- Canalis busbar
- Roxtec seals for mechanical, electrical and IT penetrations
- Hot-aisle / cold-aisle containment
- InRow cooling units
- I/O electrical distribution panels
- Novec 1230 fire suppression system
- Easy 3M UPS
- IT racks and power distribution units (PDUs)
- NetBotz monitoring
Options
- 2N power paths
- 2N UPS
- EcoStruxure DCIM (Schneider management software)
- Access control
- NetBotz environmental monitoring (Schneider environmental monitoring system)
- Automatic transfer switch
- Fire detection: VESDA
- Ventilation system
- Standalone dehumidifier
How to read the diagram: leader lines connect the equipment labels to their locations inside the central container cutaway. The four base configurations and optional equipment are listed separately on the right.
Site note, not part of the source image: the power figures are configuration labels. Their definition, GPU configuration, cooling method, installation, delivery and maintenance conditions require project-specific confirmation; no IT-load, input-power or contracted-power definition is inferred here.
BASE DESIGNS
Four base designs
20-foot ISO container
5racks / 27kWReference small AI node for low-latency inference and confidential-data management near the field.
- Factories and universities
- Hospitals and local government
- Disaster response and closed-network AI
25-foot non-ISO module
6racks / 38kWDedicated enterprise and research AI infrastructure for constrained sites, supporting progression from PoC to production.
- Enterprise private AI
- Research and development
- Expansion of existing facilities
40-foot ISO container
12racks / 80kWStandardized modular data center reference configuration for full-scale enterprise and research AI operations.
- Dedicated GPUs
- Enterprise closed-network environments
- Business continuity and industrial AI
45-foot non-ISO module
14racks / 94kWRegional distributed AI node reference configuration connecting multiple facilities, companies and local-government systems.
- Regional AI
- Ports and airports
- Disaster response and regional digital transformation
MODULAR GROWTH PATH
From field AI nodes to AI campuses
From a five-rack field AI site to a 20MW-class AI campus, the same modular approach scales with AI demand.
Reference designs of 5 to 14 racks for inference, closed-network AI, business continuity and low-latency processing at factories, universities, hospitals and local-government sites.
GPU data centers up to 2MW, considering dedicated enterprise GPUs, regional AI, high-density racks, liquid cooling, BESS and multiple modules.
Large-scale training, shared use, high-density GPUs, additional 1MW pods and multi-cluster regional AI infrastructure.
DISTRIBUTED AI VALUE
Why small-scale options matter.
Begin with the required use case: PoC, on-site inference or enterprise private AI.
Locate it where data originates: factories, hospitals, universities, ports and local-government sites.
Add modular capacity as rack, power, cooling and GPU requirements increase.
Connect AI nodes to closed networks, Regional AI Nodes and HOLONIC DATA CENTER.
INTEGRATED FACILITIES
Key equipment integrated into all-in-one designs
- 2N power paths
- 2N UPS
- Integrated DCIM
- Access control
- Environmental monitoring
- Automatic transfer switch
- Fire detection
- Fire suppression
- Ventilation
- Independent dehumidification
- Racks and PDUs
- InRow cooling
DESIGN FROM THE RIGHT SCALE
Design AI infrastructure at the scale you need.
07 / AI-READY POWER & COOLING
Power and cooling underpin AI capability before GPUs do.
AI computing resources can operate continuously only when stable power reaches high-density racks and generated heat is removed as designed.
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
Power and cooling for the GB300 NVL72 AIDC reference architecture
Design: 150 kW per rack (GB300 NVL72), liquid-cooled racks
- Cooling: direct liquid cooling (cold plates) plus in-rack CDU with N+1 pumps
- Water: warm water at 30–35°C; facility loop through a plate heat exchanger (HX) / dry cooler
- Power: MV UPS (N+1 / 2N) → LV PDU / 2N busbar → rack
- Floor loading: at least 1.6 tonnes per rack; verify structural design
- Control: BMS interlocks power, cooling and condensation protection
Power / cooling topology
6.6 kV utility supply → MV switchgear / transformer (diesel generator + BESS) → LV switchgear → UPS → LV PDU / busbar → GB300 NVL72 rack.
Dry cooler / cooling tower → plate HX → in-rack CDU → GPU cold plate (90%–98% heat removal).
Dry cooler / cooling tower → PHE → CDU → header → rack RDHx (2%–10% heat removal).
Labels in the lower power-system diagram
- Utility grid (6.6 kV) / 6.6 kV utility inlet
- Substation (MV/LV) / self-built substation
- 6.6 kV MV switchgear: VCB, protection and metering
- Main transformer: 6.6 kV / 0.4 kV, 1,500 kVA
- LV switchboard (MDB), 400 V: main bus, ACB and MCCB
- Diesel generators: 1.2 MW (N+1)
- Generator breaker and synchronization panel / ATS / STS / automatic transfer / black start
- BESS: 6 MWh battery energy storage / bidirectional PCS / EMS control / critical bus
- Data center and IT load (1 MW IT): UPS (N+1 / 2N), PDU / busway, AIDC servers (1 MW IT load)
- About 8 racks = 1 MW IT load. Rack scale with 72 GPUs and 36 CPUs, liquid cooled.
The main power path runs left to right from the utility to the data center. Diesel generators and BESS are connected below through switching, synchronization and the critical bus. GEC / GRAND ENERGY Co., LTD identifies the source.
Site note: source design figures and approximations are reproduced without recalculation. In particular, 150 kW per rack and approximately eight racks per 1 MW are separate source labels, not a newly reconciled calculation. Electrical, structural, thermal and equipment suitability requires project-specific engineering confirmation.
A simplified web summary of GEC reference designs including 150kW-class racks. It does not guarantee support for individual GPU models.
High-Density Power
- High-density GPU racks
- UPS / BESS
- Redundant power
- Modular power blocks
Liquid Cooling
- Direct-to-Chip
- CDU and warm-water loops
- Dry coolers
- Hybrid air and liquid cooling
Future-Ready
- Density changes across GPU generations
- Modular expansion
- Staged cooling changes
- Space efficiency and lower TCO
Cooling options by rack density: indicative guidance
| Rack density | Cooling option | Design considerations |
|---|---|---|
| Up to around 40kW | Air cooling | Room airflow and fan walls |
| Around 40-50kW | Close-coupled cooling | InRow, rear-door and similar options |
| Above 50kW | Consider liquid cooling | CDU and facility water loops |
| Above 100kW | Direct-to-chip and related methods | High-density heat recovery and redundancy |
Indicative only. Actual design depends on IT equipment, climate, redundancy, water-use policy and operating conditions.
08 / SOVEREIGN AI & SECURITY
Sovereign AI infrastructure to protect data, models and IP domestically.
Design not only data collection, but who may use it, for what purpose and to what extent. Closed-network configuration, storage, learning, distribution and audit form one governance framework.
Closed networksRequirements-based designs, including fully closed networks
Data separationLogical or physical separation by customer and use
Domestic storageDefine data sovereignty and storage location
Access controlManage permissions, purposes and usage scope
Audit logsPreserve operation, training and distribution history
Retention periodsLifecycle based on use and contract
AnonymizationShare only necessary information with higher layers
Model managementManage provenance, evaluation and distribution targets
BCPDesign distributed placement and continuity
Evidence preservationAddress tamper prevention and submission requirements
09 / REGIONAL AI INFRASTRUCTURE
Co-create AI infrastructure the region can use, not just a data-center location.
We envisage regional AI nodes with local government, universities, businesses and power providers, balancing everyday industrial use with continuity during disasters.
REGIONAL
AI NODE
AI NODE
Local government and universities
Factories and logistics
Agriculture and regional power
Disaster response and critical infrastructure
Disaster Resilience
River, road and slope monitoring, drone search, disaster-video analysis and regional AI operation during communications loss.
Regional Industry DX
Use field data from agriculture, livestock, factories, logistics, ports, tourism and transport within the region.
Energy Integration
Assess renewable energy, BESS, demand response, grid capacity, waste-heat use and reduced water consumption per project.
Community Coexistence
Include noise, heat, water, landscape, fuel, resident engagement, local employment and education in project design.
10 / INVESTOR VALUE
Stage investment to match AI demand and commercialization.
Instead of fixing capacity first, design power availability, anchor workloads, start-up timing and expansion units within one investment plan.
Power-secured
Integrate power capacity, connection timing, BESS and renewables.
Modular & Phaseable
Expand with demand in units such as 625kW to 2MW.
AI-ready
Plan for high-density racks, liquid cooling and future GPUs.
Time to Revenue
Aim for earlier operation through factory prefabrication and parallel construction.
Anchor Workload
Plan demand from Physical AI applications and customer workloads.
11 / STRATEGIC PARTNERSHIP
Unite Physical AI design with power, cooling and EPC.
Standard Link contributes Physical AI design and Japanese-market integration; GEC contributes power, cooling and modular EPC. These distinct specialties come together in one project design.
Workload & Japan Integration
- Physical AI workloads
- AI image analysis, cameras and Edge AI
- Drones and robotics
- IP and data governance
- Japanese-market integration and application design
Power, Cooling & Modular EPC
- Power infrastructure, UPS and BESS
- Air and liquid cooling
- Modular design
- EPCM / MEP and construction
- Operations and maintenance
- 2007
- GEC established
- Approx. 9MW
- Taiwan AI data centers
Published total IT-load track record - MODULAR
- Power, cooling and data halls
- EPC / O&M
- Design, construction, operations and maintenance
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
Power-skid deployment examples (MV and LV)
Five supplier photographs show switchgear / power-skid assemblies inside a factory, assembly work and a crane lifting a module into place. MV means medium voltage; LV means low voltage.
Source mark: GEC / GRAND ENERGY Co., LTD
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
About GEC / 碩成電能 / GRAND ENERGY Co., LTD
Founded in 2007, GEC provides supply-chain solutions for professional power-backup systems.
Building on its electrical-engineering foundation, the company has expanded into green energy, combining hardware configurations with software development to serve energy generation, storage and efficiency markets.
GEC provides comprehensive green-energy and power-integration solutions aligned with ESG principles for sustainable transformation, pursuing a long-term management philosophy that strengthens customer relationships.
Partner companies: the source states that GEC works with numerous established companies domestically and internationally.
- Logos shown: tsmc; 中興電工; Chunghwa Telecom; micron; FORMOSA PLASTICS GROUP; CTCI; STAR ENERGY; AUO / AU Optronics.
Green-energy and power-integration services
- Energy generation / energy storage / energy efficiency
- Professional power-backup system supply chain / ESG sustainable transformation
The source text states that customers are distributed across Southeast Asia and the Northern Hemisphere.
- Map labels: New Delhi, Bangalore, Singapore, Taipei, Manila, Sydney and Melbourne.
The lower-left ring connects generation, storage and efficiency around integrated services. Partner logos appear above the location map on the right.
Site note: these are GEC's source statements, not claims that Standard Link has partnerships with each named company. The source's Northern Hemisphere wording and its Australian map locations are retained as separate source information, not silently corrected.
Image text translation
Reference translation prepared for this website. The source image is unchanged; this is not an original or certified translation.
GEC history / 碩成電能 / GRAND ENERGY Co., LTD
2007–2019: building a power-technology foundation
The source identifies TSMC as the main customer, followed by deployments at large LCD-panel and memory-related factories.
- 2007: GEC established in Taipei.
- 2009–2018: GEC power-backup solutions deployed at TSMC. Other companies listed in the source include CEFC, Formosa Plastics Group and Micron Technology.
- 2019: the company established a green-energy business division and incorporated an R&D team, independently completing product, design and EMS-system R&D, testing and launch, according to the source.
2020–2024: integrated hardware and software supply chain, including technical design and patents
- 2021: independently developed the GALLANT battery brand, used in multiple applications, according to the source.
The source also reports completion of a joint energy project involving industry, government and academia together with the Industrial Technology Research Institute (ITRI) and the university named in the source as National Taipei University of Technology.
- 2023: joint development of a green supply-chain ecosystem through strategic cooperation with Schneider UPS, according to the source.
The source states that an independently developed EMS was introduced and that the company publicly entered the front-end market.
2025: energy service provider
An innovative business model for AI-powered energy services; design and implementation of AI energy solutions.
- 2025: the source states that expansion into smart-energy markets led to an official name change to Grand Energy.
An infinity-shaped band links the power-technology foundation, hardware/software integration and energy services, with the dated milestones positioned around it.
Site note: this is the supplier's historical account, not Standard Link's corporate history, partnerships or university-participation record. Names and dates are presented as source material, not new endorsements.
Third-party company names and trademarks on page 3 are part of GEC-supplied materials. All trademarks belong to their respective owners.
GEC EVOLUTION / PDF P.3–4
From power technology to integrated energy infrastructure for AI.
GEC's history and business development are summarized to clarify its role alongside Standard Link.
- FOUNDATION
Established in Taipei, building technology foundations in backup power and supply networks.
- POWER DEPLOYMENT
Developed power-infrastructure implementation capabilities through large manufacturing-site deployments.
- GREEN ENERGY
Expanded into green energy, integrating R&D and EMS.
- INTEGRATION
Brought hardware, software, design and IP into one supply framework.
- AI ENERGY
As Grand Energy, expanded into AI-enabled energy design and implementation.
These are partner achievements based on GEC and Standard Link public materials, not construction completed solely by Standard Link.
Read the strategic partnership announcement
12 / CURRENT STATUS
Where development stands today.
Track record, development, reference configurations and concepts are distinguished. Targets are not guarantees, and proposed facilities are not described as under construction or operating.
| Item | Current status | Notes |
|---|---|---|
| Strategic partnership with GEC | PARTNERSHIP ESTABLISHED | Strategic business partnership announced in April 2026 |
| Japanese-market development | IN PROGRESS | Developing projects and collaboration partners |
| AI workload design | IN DEVELOPMENT | Requirements derived from Physical AI applications |
| Modular AIDC configurations | REFERENCE ARCHITECTURE | GEC reference configurations redesigned for each project |
| First project | TARGET: FY2026 | Target for starting work; not under construction or operating |
| Regional distributed nodes | PLATFORM VISION | Concepts with local government, universities and regional businesses |
| Data Flywheel | PLATFORM DEVELOPMENT | Developing infrastructure for training, distribution and audit |
Eight months is a conditional target, PUE is a design target and specifications describe reference configurations. They are not project-specific guarantees.
13 / CO-CREATION PARTNERS
Sovereign AI for Japanese enterprises: seeking co-creation partners.
Sovereign AI means using AI while retaining data, models and IP domestically. We combine modular data-center capabilities through our strategic partnership with Taiwan-based GEC and vertically integrated technology from the edge to data infrastructure to advance sovereign AI from concept to implementation.
We seek deployment partners pursuing generative AI and AI agents in closed environments in manufacturing, finance, healthcare and local government; construction and deployment partners for land, power and construction; and sales partners for joint customer proposals.
Our policy is to become the first user ourselves by migrating our own AI development and operations to the same infrastructure.
Deployment partners
Companies and local government seeking generative AI and AI agents in closed environments
Construction and deployment partners
Companies responsible for land, power and construction
Sales partners
Companies working with us on customer proposals
13 / FAQ
Frequently asked questions.
Applications, scale, schedule, closed-network requirements, shared use and investment conditions are reviewed individually.
How does this differ from a conventional data center?
We design actual AI workloads, edge processing, closed networks, learning and model distribution together. Rather than assuming a capacity, we work backwards from what is processed and where.
At what scale can we start?
We first assess workloads and existing facilities, then size the configuration around GPU and memory, power and cooling, communications and maintenance. The 20-foot, 5-rack, 27kW design is a published reference configuration; the definition of 27kW is confirmed per project. Start small and consider expansion with usage.
Is operation within eight months guaranteed?
No. It is an operational-start target for projects with land, power, permits, procurement and construction conditions in place, where factory prefabrication and site construction run in parallel.
Can we create a dedicated closed-network AI environment?
It can be designed around requirements. We check data categories, users, connected sites, retention periods, training use and audit needs before defining the configuration.
Can local government and universities share the infrastructure?
We support joint planning for regional AI infrastructure, considering industry, research, disaster response, energy and continuity during emergencies together.
Can we discuss an investment project?
Discussions follow review of land, power, anchor workloads, SPV, EPC, operations and expansion plans.
14 / DESIGN FROM PURPOSE