Frequently asked questions

No. Coral Data Center is a thermal intelligence company. Coral researches, designs and operates the complete thermal and energy architecture of AI infrastructure across ten layers, from the silicon junction to the carbon ledger, and delivers it as modular pods customers own outright. Cooling hardware is one layer of ten, and Coral buys four of the ten layers outright rather than engineering them.

Coral maintains a continuous technical assessment of 77 companies across 18 technology segments and publishes its architecture decisions line by line, including the twenty-line bill of materials for its pods. Coral also publishes an open pod specification, an open telemetry schema and an evidence methodology that labels every number on its website by where it came from. Most modular vendors publish a specification sheet.

CoralDC is not an immersion-only company, but single-phase immersion is its current shipping architecture. CoralDC is engineering a hybrid direct-to-chip-plus-immersion platform for next-generation rack-scale systems with factory-fitted cold plates. Two-phase and passive approaches remain research directions, not production offerings.

Coral's published analysis concludes no, at the complete-system level. The compute tray ships with cold plates already fitted, a busbar carrying up to 5,000 A has no published immersion electrical safety case, the cableless board-to-board midplane is not designed for fluid ingress, co-packaged optics compatibility is unresolved, and no immersion reference architecture has been published by the vendor. Coral publishes the full component-by-component reasoning.

The leading AI silicon vendor states that its newest generation servers operate at up to 45 C coolant inlet, with coolant leaving the chips at approximately 55 C, using a coolant of 75 percent water and 25 percent propylene glycol. Coral designs its secondary loop strategy toward that specification, which enables dry heat rejection without evaporative make-up water in most climates.

No. Coral pods are designed to operate without a facility water connection in their dry rejection configuration, using dry coolers with climate-matched trim where ambient conditions require it. Conventional evaporative data centre cooling consumes on the order of 2.6 million United States gallons per megawatt per year.

The Coral OnPrem Series supports a maximum IT load of 200 kW in a 42U immersion pod measuring 2,450 x 1,050 x 1,620 mm, with a typical PUE of 1.05 to 1.15 and infrastructure overhead under 6 percent. The Coral Edge Series is a 12U pod with approximately 12 kW of usable IT load and 27 kW of combined heat rejection capability.

Coral uses two architectural answers depending on the deployment. In regulated occupied buildings Coral uses a negative-pressure water architecture, in which the entire loop operates below atmospheric pressure so a breach draws air in rather than pushing coolant out. In waterless configurations the working fluid is a non-conductive dielectric, so an escape cannot short a board. Every Coral pod also ships with oil leak detection, over pressure protection and low flow protection as standard.

Coral qualifies fluids rather than formulating them. The current standard fill is PAO or synthetic dielectric, in a design that is explicitly fluid agnostic, with a fluid capacity of 950 to 1,000 litres in the Coral OnPrem Series. Coral excludes fluorinated chemistry from every product in every market, because the category's dominant fluorinated fluid supplier exited that chemistry and regulators have designated certain related substances hazardous.

A Coral OnPrem Series pod occupies 2.57 square metres of rack footprint plus 0.84 square metres for the pump box, weighs approximately 2,200 kg operating, and imposes a floor loading of 855 kg per square metre. A typical data centre slab accepts this without modification. A typical commercial office floor requires structural review. The pod operates at ambient up to 45 C, altitude to 3,000 metres, and requires no facility water connection.

Published results from AI factory operators indicate that predictive closed-loop thermal control can release approximately 10 percent additional compute from the same power envelope by suppressing thermal spikes that otherwise force conservative setpoints. Published production results for this class of control also show 75 to 80 percent reductions in thermal spike magnitude against optimally tuned conventional control.

Coral is building toward a guaranteed Tokens per Megawatt service level agreement backed by a live digital twin and continuous telemetry, rather than a datasheet performance figure. Coral also commissions independent third-party engineering validation of its performance claims and publishes the validation in full, because sovereign and public sector procurement does not accept self-certification.

Yes. The Coral Open Pod Specification publishes the complete mechanical, thermal, electrical, telemetry, heat offtake and safety interface, together with a versioned change log. A consulting engineer can design a Coral pod into a building without a sales conversation.

Coral is building a productised heat supply product: a standard heat supply agreement, a thermal buffer that makes supply firm regardless of GPU utilisation, and a financing structure under which the heat host contributes no capital. Coral weights proximity to a heat offtaker equally with power and connectivity when selecting sites. The standardised metered heat offtake interface is a committed next-generation feature, not a current shipping one, and Coral publishes its position rather than its ambition.

Coral pods reference IEC 62368-1 and ASHRAE TC 9.9 guidelines, with CE, UL and CSA certification available as options. Coral publishes measured acoustic data, fire point and insurance classification from the fluid supplier's certificate rather than from class norms, because an insurer and a hospital estates team will both ask for the certificate.

For rack-scale integrated AI systems, immersion is not the right architecture and Coral says so publicly. For the PCIe and eight-GPU class hardware that most enterprise, sovereign, university and healthcare customers actually buy, and for board-level power delivery heat that cold plates cannot economically reach, a dielectric environment remains the best available answer. Coral's three-architecture approach exists because the correct answer differs by workload.

Established thermal vendors sell components. Coral sells an architecture, a control layer, a measured performance position and an asset the customer owns. Coral publishes its full bill of materials, its open specification and its measured data specifically so that a buyer can evaluate the engineering rather than the balance sheet, and commissions independent validation for the same reason.

Operating data from a distributed fleet of instrumented AI pods, the control layer that acts on it, and the customer relationship in sovereign and regulated markets. The consolidation of the specialist cooling sector between 2024 and 2026 transferred hardware capability to industrial majors, but it did not transfer any of those three.