Density gap
Existing rack and airflow limits may not support the target accelerator heat load.
Request a briefing ↗
01 / AI-READY RETROFIT
Modernise the live facility.
Unlock AI capacity.
IUNOVRA redesigns power, cooling, controls and heat rejection around the workload—reusing viable grid connections and assets while planning for operational continuity and creating a controlled path to high-density AI infrastructure.
DESIGN BEFORE DEPLOYMENT / TRANSFORM THE FACILITY, NOT ONLY THE EQUIPMENT
THE NEED
Reuse the site. Remove the constraints.
Many existing facilities have valuable grid access, buildings and installed assets, but cannot support the intended AI workload without upgrades. Retrofit targets the limiting power, cooling and control interfaces before a new build is considered.
Cooling constraint
Air-based systems can become inefficient or capacity-limited at higher heat densities.
Power limitation
Distribution, redundancy and controls must be redesigned around the target workload.
Live-site risk
Modernisation must protect uptime, safety and operational continuity.
THE IUNOVRA RESPONSE
One design across the facility.
Power, cooling and controls must work together. IUNOVRA defines their interfaces and operating limits before equipment is selected.
WHAT YOU RECEIVE
What the assessment delivers
- Deliverable
- A measured baseline, retrofit design and phased implementation scope.
- Site requirements
- Existing electrical capacity, cooling interfaces, equipment compatibility and live-operation constraints.
- Integration scope
- Cooling, power, controls and telemetry interfaces identified in the design.
- Acceptance criteria
- Agreed thermal, electrical and operational tests against the baseline—not a generic PUE promise.
- Support responsibilities
- Commissioning, handover, maintenance boundaries and responsibility for existing equipment.
Scope and availability
Technical descriptions include reference designs and configuration options. The proposal separates demonstrated functions, integrations to validate and development work; only the agreed scope is included. Performance and service commitments follow the acceptance criteria.
DELIVERY MODEL
From assessment to operation.
Discover
Workload, baseline and success criteria.
Survey
Site measurements and constraint mapping.
Engineer
Detailed design and implementation roadmap.
Implement
Controlled works and staged commissioning.
Validate + scale
Measured results and portfolio rollout decision.
TECHNICAL DEPTH
Retrofit is a system redesign—not a cooling equipment swap.
The engineering scope connects the workload, the live facility and the operating model. Every intervention is sized against measured constraints and commissioned against a defined baseline.
DIRECT-TO-CHIP RETROFIT
Replace the heatsink. Retain the server where compatible.
Compact sealed cold plates replace conventional processor heatsinks on compatible systems. Flexible coolant lines connect the processors to the wider cooling loop, with mounting, clearances and residual airflow validated for each server.
- Cold plates matched to processor and mounting geometry
- Flexible supply and return coolant connections
- Server-level compatibility and cooling validation
HOW RETROFIT WORKS
Capture heat at the chip.
Move it out of the building.
Cold plates are the starting point—not the whole solution. IUNOVRA integrates the server interface, cooling circuits, controls and heat rejection around the existing facility.
Explore the five cooling stages
- 01
Capture
Processor cold platesLiquid absorbs heat directly from compatible CPUs and GPUs, reducing the heat left for room air cooling.
- 02
Control
Coolant distribution unitThe CDU regulates the server cooling circuit. In a liquid-to-liquid design, a heat exchanger separates it from facility coolant.
- 03
Circulate
Pumps + distributionPumps maintain the required flow and pressure in each circuit. Distribution and redundancy are engineered for the site.
- 04
Reject
Outdoor heat rejectionA dry cooler transfers facility-loop heat to outdoor air where temperatures permit. Additional cooling depends on the climate and design.
- 05
Return
Closed cooling circuitsCooled fluid recirculates within each circuit. The server-side supply returns to the cold plates for the next cycle.
Two connected functions. Separate circuits where required. Heat passes through the CDU heat exchanger; server coolant and facility coolant do not need to mix.
Conceptual functional sequence—not a piping diagram. Equipment, temperatures and residual air cooling are validated per site.
Engineering scope and compatibility checks
Liquid-cooling architecture
Direct-to-chip loops, CDUs, manifolds, water quality, redundancy and heat rejection are engineered as one hydraulic system.
Power-path readiness
MV/LV distribution, UPS topology, busways, protection and rack delivery are checked against the target density and resilience class.
Brownfield integration
Space, structure, pipe routes, maintainability, fire strategy and live-site sequencing determine what can be reused and what must change.
Measured operating baseline
Meters, sensors and BMS/DCIM integration establish the before-state and verify PUE, thermal stability, capacity and savings after commissioning.
Transition without losing control
Method statements, change windows, rollback plans, testing and operator training reduce risk while the facility remains in service.
InfraMind optimisation
The digital twin connects asset condition, capacity, energy and workload behaviour to continuous operational decisions.
VALUE + DECISION MODEL
Four pathways—selected by the constraint, not by a headline return.
The 700-rack, 13.3 MW reference asset is a screening example. It shows how different constraints change the intervention; it is not a forecast for a specific facility.
Efficiency-first retrofit
- Keep the existing power envelope
- Illustrative PUE: 1.90 → 1.35
- Preserve rack count and footprint
Use when grid and space are constrained. The value is lower facility energy per unit of IT load and released power headroom.
Capacity expansion
- Reuse released power headroom
- Additional racks sized to validated PUE
- Demand and fit-out must be validated
Use when spare or modular space exists and there is contracted demand for additional capacity. Reusing all released headroom increases compute capacity rather than preserving the full electricity saving.
Density upgrade
- Illustrative density: 10 → 15 kW/rack
- 700 racks at full reference load
- 14.18 MW facility demand at PUE 1.35
This exceeds the original 13.3 MW envelope. Add power capacity or reduce the active rack/load requirement; higher rack density alone does not establish suitability for every AI workload.
AI-grade transformation
- Illustrative density: up to 50 kW/rack
- Major grid and electrical upgrade
- Anchor demand required before commitment
Use only where power, permits, financing and committed AI demand support a phased high-density transformation.
HOW THE CASE IS VALIDATED
Verify the baseline
Use metered IT and facility load, utilisation, tariff, occupancy, contracted pricing and the full operating-cost base—not nameplate power alone.
Engineer the delta
Model PUE and capacity improvement by phase, with redundancy, climate, load profile and commissioning tolerances.
Build the investment case
Separate gross energy savings, incremental revenue and full lifecycle costs. Confirm who pays for electricity and who retains savings under the customer contract.
Contract and measure
Tie the project to customer demand, payment milestones, M&V methodology and post-commissioning performance.
Reference sensitivity: at a constant 7 MW IT load, improving PUE from 1.90 to 1.35 releases approximately 3.85 MW and yields about €4.7M in theoretical gross annual electricity savings. A further PUE sensitivity of 1.25 releases 4.55 MW and yields about €5.6M. Both use 8,760 operating hours and €140/MWh (€0.14/kWh), before CapEx, maintenance, other operating costs, financing and tax. These are site-dependent scenarios—not EBITDA or guaranteed savings.
CUSTOMER BENEFITS
What changes for the customer.
Density
EXISTING CONSTRAINTCooling limits constrain higher-power racks.
RETROFIT OPPORTUNITYEnable higher-density workloads where power, cooling and server compatibility allow.
Efficiency
EXISTING CONSTRAINTAir cooling carries a larger share of processor heat.
RETROFIT OPPORTUNITYMove heat directly into liquid and validate the effect on PUE and operating cost.
Asset reuse
EXISTING CONSTRAINTAI upgrades can trigger a new-build discussion.
RETROFIT OPPORTUNITYRetain suitable buildings, grid connections and equipment; upgrade what constrains the workload.
Implementation
EXISTING CONSTRAINTA major replacement can create operational disruption.
RETROFIT OPPORTUNITYSequence surveys, pilots and installation in controlled phases with agreed maintenance windows.
The opportunity is site-specific. Density, savings and installation timing are established through assessment and validation—not assumed from a universal multiplier.
Retrofit improves facility readiness; it does not turn existing CPUs into AI accelerators. Server compatibility, warranty, residual air cooling and the target compute configuration must be checked. Where site constraints outweigh reuse benefits, a POD or a new facility may be the better design.
How this connects to the wider IUNOVRA platform
WHY IT IS STRATEGIC
More than a project.
A platform advantage.
Fast route to constrained capacity
Brownfield modernisation can bring capacity online faster than a new-build cycle.
Financeable efficiency
Verified energy savings may support green loans, leasing or performance-linked structures.
Platform entry point
Retrofit opens the installed base for AI POD, storage, InfraMind and managed operations.
LET’S DISCUSS YOUR PROJECT