AI Factory
delivered at speed and scale

From site planning and design to construction, IT deployment and operations. One partner integrates the facility, the hardware and the supply chain.

Talk to our experts See reference designs

Taiwan's 1st NVIDIA Cloud Partner

27 MWGB300 & Vera Rubin AI Factory

800 VDCHigh-efficiency power architecture

PUE ≤ 1.1Design target by 2028

58Dedicated AI Factory professionals

Aerial rendering of the 27 MW hybrid AI Factory campus — 27 MW Hybrid AI Factory · campus layout demonstration

Why Foxconn

Three pain points, one integrated answer

Challenge · ~~Time to deployment~~

Rapid deployment

We secure key resources and integrate vertically from the application layer to the infrastructure layer, so the AI Factory goes live faster.

Challenge · ~~End-to-end supply chains~~

Integrated ecosystem

One unified ecosystem connects and coordinates diverse AI Factory supply chains, enabling parallel execution at scale.

Challenge · ~~Cost optimization~~

Cost efficiency

Foxconn's manufacturing, design and supply chain resources work together to optimize hardware cost and TCO.

What we deliver

End-to-end, from site to tokens

Tap a stage to see what it covers.

01
EngineeringConcept to detailed design
- Concept design
- Equipment selection
- Detailed design
- Simulation & analysis
02
ProcurementContractors and long-lead equipment
- Contractor evaluation
- Interface management
- Long-lead-time equipment management
03
ConstructionPermits, civil works, facilities
- Permit application
- Civil construction
- Facilities installation
04
CommissioningL1 to L5 testing
- Commissioning testing (L1–L5)
- Construction inspection
- As-built inspection
05
IT deploymentRacks in, network up
- IT equipment deployment
- IT equipment testing
- Network bring-up
06
AI services & operationsKeep the factory producing
- AI compute service
- AI model service
- AI applications
- AI Factory operations
- Maintenance service

Build types

Choose the build that fits your time to market

Estimated lead time for a 10 MW AI DC, in months

Standard

36–48

Container

12–18

Hybrid

12–18

Hybrid DC (HDC)

12–18 months10 MW estimate

Balance between customizability and rapid deployment
IT equipment in a steel-structure building
Modular cooling and power equipment

Vertical integration

From components to the whole factory

Foxconn builds every layer, so design, supply and delivery stay in one hand.

Components

L3 · L5 · L6

Cold plateManifoldQDBusbarPower modulePCB / PCBAEnclosureChassis

AI server

L10 · L11

Compute traySwitch trayStorage trayRack

AI DC ecosystem

L12+

ITHVACPowerStructure & auxiliary

Design roadmap

Toward PUE ≤ 1.10 and near-zero water

Cooling and power evolve with each NVIDIA generation to raise tokens per watt.

2026GB300

< 1.30PUE

87%Liquid cooled

—WUE

BaselineCooling energy

Power: 480 VAC

2027Vera Rubin

< 1.20PUE

96%Liquid cooled

< 1WUE

−15%Cooling energy vs 2026

Power: HVDC 800 V with power center

2028Vera Rubin next-gen

< 1.10PUE

TBCLiquid cooled

≈ 0WUE

−30%Cooling energy vs 2026

Power: HVDC 800 V with SST

2027 and 2028 figures are design targets. Cooling energy is compared with our 2026 GB300 design.

Digital twin

Built virtually first with NVIDIA^® Omniverse DSX^™

Our 27 MW AI Factory has a full digital twin for engineering validation and deployment.

Digital twin view of the Foxconn 27 MW data center

01Simulate & predictPhysics simulation with DSX SimReady finds risks before deployment.

02OptimizeReal-time PUE with DSX Flex tunes performance and efficiency.

03ImproveOT monitoring validates results and feeds the next cycle.

A closed loop for maximum tokens per watt

Reference designs

Start from a proven unit, then scale

Next-gen hybrid AI Factory: steel-structure IT halls with modular power and cooling.

Spec	27 MW	3.3 MW
IT load (incl. network)	22 MW	2.8 MW
GB300 racks	144 16 per POD	16
GPUs	10,368	1,152
Architecture	9 ComputePOD + 1 CorePOD	1 SuperPOD
Area	8,232 m² 147 × 56 m data hall	4,183 m² 89 × 47 m total
Power module	5 MW	0.5 MW

Uptime Tier III (N+1)Estimated PUE ≤ 1.200.136 MW per GB300 rack72 GPUs per rack

Global strategic partners

A proven ecosystem behind every build

IT

NVIDIA · Dell · HPE · Gigabyte · Arista · Cisco · Oracle · NetApp · WEKA

HVAC

Vertiv · Schneider Electric · Johnson Controls · YORK · SPX · Trane · Munters · Carrier

Power

ABB · Delta · Eaton · Siemens · Caterpillar · Kohler · TECO

EPC & commissioning

Exyte · WSP · Uptime Institute · TECO

Let's discuss your AI Factory project

Share a few details and our AI DC team will review your inquiry and get in touch.

AI Factory
delivered at speed and scale