Skip to content

Green DC embeds AI compute in the local energy system

How an AI compute load fits into an energy system that already exists: what it consumes, what it returns, and what it asks of the site that hosts it.

Write to us

The structural gap

Why the existing base will not be enough

Demand for AI compute is growing fast and pushing rack power densities beyond what most existing data centres can handle. AI clusters require at least 40 kW per rack and water cooling rather than air.

Training is geographically flexible; inference remains latency-sensitive. That is what makes edge deployment relevant, complementing the hyperscalers rather than replacing them.

These constraints reveal a structural gap between the new requirements of AI infrastructure and traditional data centre deployment models.

L'architecture de référence

Two energy sources, and heat that is reused

The facility fits into an energy system that already exists. It draws locally generated energy as its primary supply, returns the waste heat the compute produces, and keeps the public grid as a redundancy layer.

01

Local generation is the primary supply

Whatever the technology, provided it delivers a continuously usable profile: dispatchable by nature, or firmed by storage.

02

Waste heat is recovered as a rule

Fed back into district heating, an industrial drying process, a horticultural greenhouse or a public facility.

03

A resilience layer covers continuity

Grid interconnection, storage, backup generators: the architecture is grid-light, not off-grid.

Dual energy sourcing by design. Local energy is the primary supply; the public grid provides redundancy. Two sources where a conventional data centre has one, with grid demand limited to cover rather than supply.

Reference architecture

Local energy generation, primary

RDF cogeneration
Biomass cogeneration
Solar PV + BESS
Wind + BESS

Modular AI data centre

High density · direct liquid cooling

Resilience layer

Public grid as backup, grid-light architecture
Storage and backup generators
Two fibre routes on separate paths

Output

AI compute capacity

Waste heat recovered

Grid-light, not off-grid: the public grid stays connected as redundancy. The GDC 01 pilot uses an RDF cogeneration configuration.

Eligible energy sources

Two families, one criterion: a usable 24/7 profile

We have no technology preference. The criterion is not how the energy is generated, but the ability to deliver continuous power and the existence of an outlet for the heat.

Family 1

Dispatchable generation

RDF or biomass cogeneration. Output is continuous by nature: it depends on neither wind nor sunshine, and its profile maps directly onto a compute load.

Cogeneration from refuse-derived fuel
Biomass cogeneration

Family 2

Renewables firmed by storage

Solar or onshore wind paired with battery storage. Output is intermittent, but storage smooths it: the generation + BESS pair delivers the usable profile, never the farm on its own.

Solar PV + BESS
Onshore wind + BESS

The design

Modular, dense, replicable

Modularity

Industrialised units deployed incrementally. Each module carries standalone infrastructure matched to one generation of compute processors.

AI-grade density

Rack density and cooling sized for AI loads from the outset, without the structural limits of legacy air-cooled data centres.

Replication

Standardised yet adaptable design, shared engineering, centralised intellectual property and controlled deployment.

Industrial anchoring

Each site remains tied to a specific asset and to its local energy, land and permitting constraints. That is the part that does not replicate, and we say so.

Distributed resilience

Several mid-sized sites rather than a single large one: geographic concentration risk falls.

Low-carbon construction

Units designed for low-carbon construction, including timber structures where relevant, using local industrial capacity.

Energy efficiency

Built in, not merely declared

Cooling

Latest-generation technologies, including direct liquid cooling, to enable high-density loads and materially reduce PUE.

PUE target

To outperform new-generation benchmarks of below 1.33 by a wide margin, thanks to colocation with a local generation unit.

Water

Very low consumption by design: the loop is closed. The project does not compete with local uses.

Waste heat

Recovered for local industrial uses or district heating. Integration with a cogeneration asset improves overall efficiency.

Regulatory alignment

A measurable approach, aligned with sustainability reporting requirements and eligible for public support for energy optimisation and industrial decarbonisation.

PUE: power usage effectiveness. New-generation benchmark: PUE below 1.33.

Deployment

A pilot, then a model

GDC 01, in south-eastern France, is the demonstrator: RDF cogeneration configuration, over 3 MW IT modular. It validates the architecture, the energy integration and the operating model.

Subsequent sites reuse the validated design: colocation with existing energy units, fast deployment, low dependence on the electricity grid.

Let's talk about your project

The three direct addresses are on the contact page.

Write to us