Fit the site you have
Layout, utility connections and access are developed around your location and available power.
THE FACILITY FOUNDATION FOR DEPLOYABLE COMPUTE.
MODFLEX MODULAR DATA CENTERS
Modular data centers. Built around your project.
ModFlex brings the building, power and cooling together for AI and HPC. Start with the capacity you need, on the site you have, with a clear path to expand.
THE MODFLEX APPROACH
Your site. Your capacity. Room to grow.
A purpose-built modular building system, not a fixed shipping-container catalogue. We bring the building, power, cooling and operating requirements into one facility design, with expansion planned from the start.
Support the compute you need
Rack loads guide the power distribution, air or liquid cooling, and service access.
Add capacity in stages
Plan the first deployment and the provisions for future growth before committing to the layout.
What goes into your facility
Site and utility interface
Plot constraints, utility connections, access, permitting inputs and the route from available power to usable IT capacity.
Power architecture
Load profile, resilience targets, distribution, protection, metering and planned growth considered as one electrical path.
Cooling and heat rejection
Air, direct-liquid or hybrid cooling selected around actual rack loads, climate, water strategy and service requirements.
Capacity and expansion
The first deployment and future physical, electrical and hydraulic provisions defined before the layout is fixed.
Operational fit
Service routes, controls, security boundaries and support spaces configured around the way the facility will be run.
Building modules and site integration
Engineered structural modules
Steel-frame building modules developed around equipment loads, transport, assembly, durability and site requirements.
White space and support areas
Rack halls, network and storage zones, electrical rooms, cooling plant, staging and operational spaces placed around the intended deployment.
Architectural and site integration
External finishes, dimensions, circulation, security boundaries and surrounding infrastructure adapted to the location and planning context.
Repeatable expansion
Defined growth stages let additional capacity follow demand, utility availability and commercial timing.
FROM FIRST CONVERSATION TO HANDOVER
One coordinated delivery path.
Site work, engineering, manufacture and commissioning follow the same capacity plan. Scope, schedule and acceptance criteria are agreed for your project.
- 01
Baseline
Confirm the site, power position, workload, target capacity, timeline, resilience objective and operating model.
- 02
Concept design
Translate those inputs into module layout, capacity stages, rack assumptions, electrical architecture, cooling strategy and budget basis.
- 03
Detailed engineering and procurement
Coordinate structural, electrical, mechanical, controls and equipment packages against approved requirements and interfaces.
- 04
Manufacture and site integration
Progress module manufacture while site preparation, utility work and deployment logistics move forward in parallel.
- 05
Commissioning and handover
Test systems, controls, failure responses and operating procedures against the agreed commissioning and acceptance plan.
PLATFORM-AWARE FACILITY DESIGN
Your facility. Your choice of compute.
ModFlex can house your selected IT systems, or ElioVP can help integrate the compute, networking and storage. Facility design follows confirmed rack requirements; future platforms remain planning scenarios until their configurations are confirmed.
Explore AI infrastructurePlatform requirements and operational access
Current AI platforms
Plan around confirmed AMD MI355X, NVIDIA B300 and NVIDIA GB300 NVL72 system requirements and selected OEM rack data.
2026 platform rollout readiness
Model electrical, hydraulic, structural and spatial pathways for AMD Helios partner systems and NVIDIA Vera Rubin deployments, then confirm every assumption against the selected configuration.
Network and storage zones
Include fabric switches, storage systems, management networks, cable pathways, service clearances and supporting racks in the capacity model.
Agreed availability target
Redundancy, maintainability and failure-domain decisions follow the project’s operational requirement rather than a blanket claim.
Operational access
Equipment replacement, isolation, maintenance routes, controls and safe working access are considered before the layout is fixed.
Integrated commissioning
Electrical, mechanical, controls and operational scenarios are tested against a project-specific commissioning plan before handover.
ENGINEERING AND TRUST
Know what sits behind the design.
Tier Standards expertise
Our engineering team includes a Uptime Institute Accredited Tier Specialist. That expertise informs decisions around resilience, availability and operations, but does not imply a project is Uptime Institute certified unless certification is separately scoped and achieved.
Private projects. Verifiable experience.
We respect our customers’ privacy and take sovereignty and operational security seriously. Many completed environments cannot be published publicly, but selected project photography, video and technical references may be shared privately with qualified organizations where customer permissions allow.
Request private project referencesLOOK INSIDE MODFLEX
Explore the engineering, at your own pace.
Open the areas relevant to your project. These are separate illustrative configurations and studies, not one project specification or a guarantee of performance.
01Explore the modular designInspect the structure, rack layout, power and cooling in an interactive 3D example.Configurable architectural reference
MODFLEX / INTERACTIVE EXPLORER
Your datacenter. Your design.
Fully modular, from the floor plan to the facade. AMD Helios is our flagship showcase. ModFlex can also be designed around NVIDIA and other AI, HPC or storage platforms, with the layout, systems and finish shaped around you.
Compute platform
The architecture shapes the space, too.
8 Helios · 6 storage · OOB
Exterior finish
A few possibilities. Specify your own facade.
A closer look
01 /
Fully modular. Entirely yours.
This is one possible configuration. Arrange the white space, Powerpods and cooling around your site and workloads. Choose the architecture, specify the finish and plan how you grow — with no fixed catalogue layout.
Freedom by designYour layout. Your capacity. Your architecture.
System layers
Flagship showcase: AMD Helios. ModFlex modules can be configured for other compute platforms. Layout, equipment, electrical sizing and cooling are tailored to each project. AMD Helios reference
One example. Your possibilities.
Tell us how you want to build. We’ll shape the modules, exterior and infrastructure around you.
02See how cooling is studiedFollow the airflow and see how simulation informs the cooling design.CFD study · 12 RDHx racks · 600 kW reference load
MODFLEX / AIRFLOW ENGINEERING
See the airflow.Refine the design.
Good cooling design goes beyond choosing equipment. We use computational fluid dynamics (CFD) to examine how air moves through the room, where warmer air travels and how the layout shapes the result.
Compare the assumptions. Inspect the results. Document the design decisions.
INSIDE A MODFLEX DESIGN STUDYCFD / 01
22.5 seconds · Silent film
01AIRFLOW
Follow the air.
Trace calculated circulation through the rack layout and rear-door heat exchangers (RDHx), with the room dimensions in view.
12 RDHx racks · 600 kW planned IT load
02TEMPERATURE
Test the assumptions.
Set one rear-door discharge to 38°C and examine where warmer air travels. A deliberate sensitivity case, not a recorded equipment failure.
38°C prescribed input · 32°C review threshold
03PRESSURE
Inspect the differences.
Examine calculated pressure differences across the room. Read each section against its legend and the defined scenario.
Relative room pressure, not fan pressure rise
What this study shows, and what it does not
A defined model
The airflow and pressure study uses a 99,960-cell model of a 13.416 × 4.700 × 2.836 m room, with 12 RDHx racks, 600 kW planned IT load and 122,400 m³/h prescribed recirculation. These are scenario inputs, not a certified capacity rating.
A separate thermal case
The temperature study uses 9,480 cells, with RDHx 06 discharge prescribed at 38°C and the other doors at 29°C. The calculated maximum is 38.0°C; 14.3 m³ of room air exceeds that case’s 32°C review threshold. The discharge temperature is an input, not a predicted coil outlet temperature.
The right interpretation
These are preliminary steady-state CFD results. Moving particles illustrate the solved velocity field, not a transient simulation. Pressure is relative to the volume-weighted room mean, with hydrostatic head removed. The results do not establish fan pressure rise, coil resistance, acoustics or validated cooling performance.
Your layout. Your loads. A considered cooling design.
Bring your rack plan and operating assumptions. We can explore the scenarios that matter to your project and preserve the findings in an engineering report.
03Explore power and cooling failuresCut the grid or stop a pump. See how backup systems and remaining capacity respond.Resilience example · 16 GB300 NVL72 racks · 2.40 MW IT load
MODFLEX / INTERACTIVE RESILIENCE LAB
What happens whensomething stops?
Cut the grid. Trip a pump. Take a power pod out of service. Explore how backup systems take over and how the remaining capacity responds, before a real failure ever happens.
Follow each change from the failed component to the rack. Try a prepared scenario, or build your own combination.
ILLUSTRATIVE REFERENCE BLOCK16 × GB300 NVL72 racks
POWER / ONE MODFLEX BLOCK
Four independent paths. One continuous supply.
Cut the grid and follow how UPS batteries bridge the load while generators start. Click a component to simulate a fault or restore it.
Nominal operationAll four paths are supplying power.
Live paths4 / 4
8 / 8 rack feeds available
Power per rack150 kW
150 kW requested
Available compute100%
2.40 MW · Block IT load
Simulation time00:00
1× · Simulation
Choose a scenario
Each scenario starts afresh with fully charged batteries.
Swipe sideways to explore the full diagram. Every component is also available under Advanced controls.
RACK R011 of 16 identical racks
PS1 Path A18.8 kW
PS2 Path A18.8 kW
PS3 Path B18.8 kW
PS4 Path B18.8 kW
PS5 Path C18.8 kW
PS6 Path C18.8 kW
PS7 Path D18.8 kW
PS8 Path D18.8 kW
Generator + ATS · 10 s · 16 racks · 8 × 33 kW
60 kW190 kW
At 150 kW, three paths carry the full rack load. With two paths, up to 132 kW per rack remains available.
Simulation
The clock pauses automatically when this diagram is out of view.
Advanced controls
Click to simulate a fault. Click again to restore the component.
Path A
Path B
Path C
Path D
What happens next?
1- Block energised: all four paths live.
Illustrative model, not live telemetry or an availability guarantee. 16 racks, 33 kW per power shelf, four independent paths and an assumed 10-second generator start. Actual performance depends on the build and project design.
COOLING / SHARED HEADERS
Take a cooling train out. Follow the water.
All cooling trains available
Four CDUs, four pumps and four dry coolers share the load.
Available compute100%
Supply to racks35.0°C
Rack pressure difference1.20bar
Liquid-cooled heat1.92MW
Swipe the diagram sideways. Click a CDU, pump or dry cooler to trip or restore it. The same controls are available below.
Rack supplyFacility supplyDashed: 35 °C setpoint
Rack differential pressureDashed: 1.2 bar design point
Start with a scenario
Change the conditions
100% represents 1.92 MW of liquid-cooled heat. Adiabatic pre-cooling assists above 28 °C ambient.
Control each component
Shared headers let the remaining three components in each family carry the illustrative design load. Click to trip or restore.
| Train | CDU | Facility pump | Dry cooler |
|---|---|---|---|
| A | |||
| B | |||
| C | |||
| D |
What the model is doing
- Cooling at the design point. All trains running.
00:00 simulated
Illustrative concept inputs: 35/45 °C rack water, 30/40 °C facility water and 1.2 bar design pressure difference. Actual behaviour depends on the engineered installation, operating conditions and controls.
What this example models
A separate reference block
This example uses 16 GB300 NVL72 racks at 150 kW per rack: 2.40 MW of IT load, with a 1.92 MW liquid-cooling share. It is separate from the configurable 3D layouts above. It does not model every facility load or the remaining air-cooled heat.
Four electrical paths
Each path has a grid feed, generator, UPS, battery and busway, with two 33 kW power shelves per rack. The example generator start takes 10 seconds. Battery endurance varies with load; loss of multiple paths can force rack power capping.
Shared cooling headers
Four CDUs, four facility pumps and four dry coolers serve shared headers. The design point is 25 °C ambient, with illustrative 35 °C rack-water supply and 1.2 bar rack differential pressure. Capacity depends on load, remaining equipment and ambient conditions; adiabatic assistance is assumed above 28 °C.
A way to explore, not a guarantee
Transfer times, thermal inertia, temperatures and compute percentages are simplified concept assumptions. These are not live readings, a validated digital twin, a Tier certification or an availability promise. Every ModFlex project requires its own electrical, hydraulic, controls and commissioning studies.
Design around the failures that matter.
Tell us your availability target, rack loads and maintenance requirements. We will define the right failure scenarios and commissioning plan for your site.
LET’S PLAN YOUR MODFLEX PROJECT
Start with your site and what you need to run.
Share what you know so far. We will help turn it into a facility scope, capacity plan and practical next steps.
- Location
- Available power
- Workload & capacity
- Target timeline