Reference architectures
Published designs for situations CoralDC encounters repeatedly. Each describes the constraint, the architecture, the thermal and electrical approach, and the conditions under which it does not apply. These are engineering documents rather than marketing collateral, and they exist partly because they are the strongest proof a company can publish before customers will go on record.

AI infrastructure in a hospital without a data hall
The constraint. No data hall and no room to build one. Constrained electrical capacity in a building that has been extended repeatedly. Floor loading limits in older wings. Acoustic limits near clinical areas. Infection control requirements that restrict what can be done to air handling. Patient data that cannot leave the institution. And a leak risk question that, above an operating theatre, an imaging suite or a pharmacy, is frequently a veto rather than a risk assessment.
The architecture. A sealed, liquid-cooled pod installed in existing plant space, a service area or a repurposed room, sized where possible to operate within the building's existing electrical service. Sealed thermal path so no air handling modification is required and infection control is not engaged. Fans removed from the thermal path, which addresses the acoustic constraint directly. Outdoor cooling unit sited on an existing plant deck or roof with a short pipe run.
Thermal approach. Single-phase immersion today. Leak risk is treated as a primary design constraint rather than a secondary one, which is why CoralDC also evaluates negative-pressure water architectures, where a breach draws air inward rather than expelling fluid, for the direct-to-chip half of the future hybrid platform.
Electrical approach. Sized to existing capacity wherever the density target permits, with distribution, protection coordination and monitoring engineered at the building interface.
Where this does not apply. Buildings where structural capacity at any candidate location is below the operating weight, where no plant location exists within an economic pipe run of the pod, or where the density target genuinely cannot be met inside the available electrical service.

High-density compute at a remote industrial site
The constraint. Airborne dust at concentrations that destroy air-cooled equipment. Vibration. Extreme ambient temperature range. Humidity. Frequently no building suitable for IT equipment, no on-site technical staff, and satellite or constrained terrestrial connectivity that makes moving raw data off site impractical for anything time-sensitive.
The architecture. A sealed pod with the thermal path closed to the environment, remote monitoring and managed operations, and an operating envelope that accommodates the ambient range rather than requiring a conditioned room to be built around it. Operating temperature range minus 25 C to plus 50 C, relative humidity 5 to 95 percent non-condensing, altitude to 3,000 metres.
Thermal approach. Single-phase immersion. This is the deployment class where sealing the thermal path is not an efficiency argument but a survival one: it removes airborne contamination as a failure mode entirely, which in a processing environment is the difference between equipment that runs for years and equipment that fails in months.
Where this does not apply. Sites where the ambient range exceeds the published operating envelope, where delivery access cannot accommodate the pod, or where no serviceable power source exists.

University research computing under data residency requirements
The constraint. Many principal investigators sharing capacity from a budget that is rarely sufficient. Wildly variable and bursty workloads. Funder and ethics conditions that prevent some datasets leaving the institution. Chargeback obligations. And an estate that is frequently the oldest and most constrained on campus.
The architecture. A pod sited in existing campus plant or service space, with multi-tenancy and project isolation through Coral Cloud so many groups share capacity safely, access control integrated with institutional identity systems, and per-project utilisation reporting for chargeback. Data remains within the institution and its jurisdiction.
Thermal approach. Single-phase immersion, with the acoustic and air handling advantages that matter in buildings shared with teaching and office space.
Where this does not apply. Institutions whose funder conditions require a specific accredited facility type, or where the workload profile genuinely suits burst cloud capacity better than owned infrastructure.

Government departmental AI under jurisdictional control
The constraint. Jurisdictional requirements on data location. Security requirements on physical and logical access. Supply chain assurance obligations. Procurement rules requiring demonstrable domestic capability. And an institutional reluctance, frequently well founded, to accept a dependency that cannot be reversed.
The architecture. Infrastructure on premises the department specifies, under its physical control, with operations retained or delegated reversibly, and a documented architecture designed to remain operable and serviceable independent of CoralDC.
Thermal approach. Single-phase immersion, sealed, with no air handling dependency on the host building.
Where this does not apply. Environments requiring a security clearance level or controlled goods registration that CoralDC does not currently hold. CoralDC states that position plainly on the Defence and Procurement pages rather than implying otherwise.
Request an Architecture Review
A structured technical conversation about power, thermal capacity, density, jurisdiction and deployment sequencing for a specific site. If the answer is that CoralDC is not the right fit, it says so.
