FAQ

Although Éclairion is not a data hosting provider, the sovereign location of the physical layer remains a regulatory issue for its clients. GDPR and the French State Cloud doctrine govern not only data itself, but also the conditions of physical access to the infrastructure that supports it. An OIV, healthcare institution or public administration installing its equipment in an AI Factory must be able to guarantee to its supervisory authorities that the physical layer is beyond the reach of any foreign jurisdiction — including in the event of legal proceedings. Éclairion infrastructure meets this requirement. The SecNumCloud qualification, which our cloud operator clients may themselves target, relies notably on control of the physical layer — a layer that Éclairion provides them.

The development of sovereign AI infrastructure in France is aligned with the priorities carried by the France 2030 plan, which identifies computing power and AI infrastructure as national strategic axes. At European level, IPCEI (Important Projects of Common European Interest) programmes in the cloud and AI infrastructure domain fund projects of common interest dimension — of which high-performance AI Factories form part. This convergence between market dynamics and public investment priorities reinforces the relevance and durability of the Éclairion model beyond its purely commercial logic.

Éclairion's sovereignty is that of the physical layer — and that is precisely where it matters most. A foreign AI Factory operator providing space, power and cooling in France would remain subject to its national legislation of origin, which could compel it to grant access to its facilities, its technical logs or its physical data flows. Éclairion is a French company, owned by exclusively French and European capital, operating solely under French and European law. Its sites, buildings, electrical and cooling systems are French assets. For a client installing its GPU clusters there, this guarantees that the physical layer of its AI infrastructure depends on no foreign jurisdiction.

Éclairion's environmental commitments are measured, audited and certified — not declarative. Energy management is certified ISO 50001, audited annually by an independent body. Consumption and efficiency indicators (PUE, WUE, valorised heat) are made available to clients for their own reporting. On biodiversity and land use, Éclairion commits to verifiable site-by-site metrics: greened surfaces, ecological continuities created, volumes of water managed by natural infiltration. Sustainability is not a communications angle at Éclairion — it is a design constraint integrated from the first line of the architect's brief.

Direct Liquid Cooling (DLC) is a family of cooling technologies in which a heat-transfer fluid is brought physically into contact — or in immediate proximity — with the components that generate heat, rather than cooling the ambient air around the servers. This is a fundamental break from conventional air conditioning, which becomes inefficient beyond a certain thermal density threshold. A high-performance GPU rack can dissipate between 50 and 100 kW of heat: air is no longer sufficient to evacuate it without consuming disproportionate amounts of energy. DLC solves this problem at source — and that is why any serious high-density AI Factory deploys it today. Éclairion integrates DLC from the architectural design stage of its sites, not as an afterthought.

These are two variations of DLC, adapted to different contexts. Immersion cooling involves submerging servers — motherboards, GPUs, memory and all — in a bath of dielectric fluid (an electrically insulating fluid, safe for electronic components). Heat is absorbed directly by the liquid, with no air intermediary. This is the most thermodynamically efficient solution: the heat transfer coefficient of a liquid is hundreds of times greater than that of air. It enables extreme rack densities and recovers heat at high temperature, facilitating its valorisation. Direct-to-chip (or cold plate) is a more selective approach: a coolant circuit circulates through metal plates fixed directly onto the hottest components — typically GPUs and CPUs — while the rest of the server remains air-cooled. It is a compromise between thermal performance and compatibility with standard equipment. Éclairion deploys both technologies according to the needs and configuration of each site, systematically prioritising the most efficient solution for the targeted density.

France has one of the most decarbonised electricity grids in Europe, with more than 70% of electricity from nuclear origin. For an AI Factory consuming several tens of megawatts, the national electricity mix is not a detail: it directly determines the environmental footprint of each computation performed. This is a structural and lasting advantage — rooted in the French production fleet, independent of market fluctuations. Éclairion secures its supply through dedicated energy purchase contracts, site by site, guaranteeing multi-year price stability and traceability of the origin of electricity supplied to each client.

Éclairion does not establish on a territory to consume space and energy — it establishes there to sustainably revitalise it. Each AI Factory generates permanent jobs: operations technicians, engineers, support functions, but also the full range of induced activities in the local fabric — maintenance, services, specialist subcontracting. These jobs are anchored in the territory, non-relocatable, and span the lifetime of the infrastructure. Beyond employment, each site represents a significant investment for the host commune and region: local tax revenues generated, orders placed with basin businesses during construction and operation, and rehabilitation of a brownfield that weighed on the territory's attractiveness. An Éclairion site transforms a land liability into a lasting local economic asset.

Éclairion sites are selected according to three non-negotiable criteria: electrical connection capacity (an AI Factory site consumes several tens of megawatts — only high-voltage network nodes can supply it), brownfield land availability (we do not build on virgin land), and geographical distribution (our six sites cover several zones of the territory to enable redundancy and user proximity). Among these sites is Bruyères-le-Châtel (Essonne, Île-de-France) — a historically industrial and scientific territory that already hosts significant national-scale intensive computing infrastructure, with connection capacities adapted to high-density GPU loads.

Beyond brownfield rehabilitation, Éclairion imposes a discipline of active ecological compensation on each of its sites: greening of non-built spaces, creation or restoration of ecological corridors, partial de-sealing of surfaces where technically possible. The goal is to return to nature a surface at least equivalent to the net built footprint of each AI Factory. This involves concrete architectural choices — green roofs, stormwater management by natural infiltration, considered landscape integration — and partnerships with local biodiversity actors to monitor the evolution of sites over time.

The choice of brownfields is both a conviction and a discipline. Building a new AI Factory on virgin or agricultural land irreversibly consumes natural spaces, fragments ecosystems and contributes to soil artificialisation — a phenomenon France has committed to drastically reduce (zero net artificialisation target by 2050). Industrial brownfields offer an alternative: already sealed soils, often well connected to existing electricity networks (former factories, heavy industrial sites), and territories that directly benefit from rehabilitation. By establishing on these sites, Éclairion restores a productive purpose to abandoned spaces, creates local jobs and avoids encroaching on agricultural land or natural areas.

The high availability of an AI Factory rests on three pillars: electrical redundancy (dual power supply, generators, UPS dimensioned to maintain GPU load without interruption), robustness of cooling systems (circuit redundancy, continuous thermal monitoring of racks), and 24/7 supervision of physical infrastructure. Éclairion deploys architectures compliant with high-availability datacenter Tier standards. Furthermore, its network of six geographically distributed sites allows clients to replicate their workloads across multiple sites simultaneously, eliminating any single point of failure.

In the high-density colocation model, the client organisation brings its own GPU servers — whether NVIDIA H100, B200 or other architectures — and installs them in Éclairion racks. It benefits from power, cooling and connectivity provided by Éclairion, but remains owner and operator of its hardware. This model suits players who want complete control of their stack and full independence from any service provider. The Infrastructure as a Service (IaaS) model means GPU computing capacity is provided without the client acquiring hardware themselves — Éclairion or partners operate the equipment according to contractually defined performance and availability commitments.

An AI Factory is a physical infrastructure designed end-to-end to produce artificial intelligence at scale. Training a large language model, running clusters of thousands of GPUs in parallel or powering large-scale inference platforms requires electrical and thermal densities far beyond conventional enterprise servers: a high-performance GPU rack can consume between 30 and 100 kW, compared to 5–10 kW for a standard rack. Buildings, electrical systems and cooling solutions must be dimensioned accordingly from design stage — a retrofit is neither economical nor effective. This is exactly what Éclairion builds: buildings natively designed for high-density AI computing, on sites chosen for their electrical connection capacity.

Éclairion addresses any organisation needing intensive computing infrastructure in France: AI model publishers, large industrial groups developing their own AI capabilities, sovereign cloud operators seeking top-tier physical hosting, public or private research laboratories, and integrators building GPU-as-a-service platforms for their own clients. The common requirement: computing power and electrical density needs that exceed what a generalist datacenter can offer, with a requirement for French territory location.

Éclairion provides three things, and three things only: space (high-density racks in a secure building dimensioned for extreme GPU thermal loads), electrical power (redundant, high-availability supply with direct national grid connection) and cooling (advanced systems — immersion, direct-to-chip — capable of handling AI infrastructure thermal densities). This SPC (Space, Power, Cooling) model is used by major global infrastructure operators. It guarantees each client total sovereignty over their equipment: Éclairion owns the building and fluids, not what happens inside.

An AI Factory is a physical infrastructure designed end-to-end to host the most intensive GPU computing workloads — training large AI models, running clusters of thousands of GPUs in parallel, or powering large-scale inference platforms. Éclairion is its operator: it provides the built space, electrical power and cooling (Space, Power, Cooling) required for these extreme workloads. Its clients bring their own equipment, deploy their own systems and retain full control over their data and models. This is a fundamental difference from a conventional datacenter: Éclairion provides the physical layer — and exclusively that. What clients do with this computing power is entirely their own business.