IN Brief:
- Frore Systems announced LiquidJet Diamond on 7 October, incorporating diamond heat spreading into a coldplate.
- The design builds on the company's three-dimensional, short-loop, multi-stage liquid cooling architecture.
- Its claim of up to 35% higher AI factory profitability is a manufacturer-modelled comparison.
Frore Systems has introduced LiquidJet Diamond, a liquid cooling coldplate that incorporates a diamond heat spreader to address concentrated heating in high-power AI processors. The company announced the design on 7 October as an extension of its existing LiquidJet technology.
The development targets thermal hotspots within accelerator packages, where the power dissipated by individual regions of a semiconductor die can be considerably greater than the average heat flux across the package. A cooling arrangement that handles the average heat load may still allow local temperatures to approach operating limits.
Liquid cooling removes heat by conducting energy from the processor through mechanical and thermal interfaces into a coldplate, then transferring it to circulating fluid. Performance is influenced by the resistance of each interface, the area available for heat transfer, the coolant conditions and the distribution of fluid through the plate.
Diamond is an attractive heat-spreading material because it can conduct heat rapidly under suitable material and assembly conditions. Spreading energy laterally can lower a concentrated temperature peak before that heat reaches the liquid cooling structure. The resulting improvement depends on the geometry and quality of the interfaces as well as the nominal conductivity of the diamond.
A highly conductive spreader cannot remove every source of thermal resistance. The processor package, interface compounds, mechanical contact and liquid-side heat transfer still influence the difference between semiconductor junction temperature and coolant temperature. Characterising the complete assembly is consequently necessary to evaluate the benefit.
Frore describes its underlying LiquidJet design as a three-dimensional, ultra-short-loop, multi-stage liquid cooling structure. The company says it can be customised to the power map of a GPU, placing the cooling capacity where the processor releases the greatest amount of heat.
Tailoring a coldplate to a particular processor can make sense because accelerator architectures have different die arrangements and hotspot distributions. A design optimised for one package cannot automatically be assumed to reproduce the same result on another device with different chiplets, heat spreaders or thermal interface materials.
Frore claims the LiquidJet Diamond approach can increase AI factory profitability by up to 35% compared with traditional skived coldplates. In its 7 October announcement, the company links this figure to its model of processor cooling, AI tokens per watt and potential revenue from computing output.
This percentage is not a measured profit increase demonstrated across deployed data centres. It is a manufacturer comparison whose outcome depends on assumptions about the accelerator, operating temperature, workload, utilisation, electricity cost and how computing services generate revenue. Those assumptions must be considered before applying the percentage to a particular facility.
Better cooling can increase performance headroom where thermal limits would otherwise cause a processor to reduce its operating frequency or power. The size of that effect depends on control behaviour and workload, so a measured temperature reduction does not necessarily produce an equivalent increase in useful computing throughput.
A liquid cooling system also consumes energy outside the semiconductor package. Pumps must overcome the pressure drop of the fluid circuit, and the heat removed from the processor must ultimately be rejected elsewhere. Overall efficiency therefore requires evaluation of the coldplate, pumping, distribution and facility heat-rejection arrangements together.
Liquid flow geometry creates its own trade-offs. Small channels can increase local heat-transfer effectiveness but may also raise hydraulic resistance, making pump selection and coolant management important. System designers must consider filtration, material compatibility and the maintenance conditions under which the equipment will operate.
The announcement does not give an independently validated comparison of full-facility power use or a public customer deployment dataset. The confirmed October milestone is the integration of a diamond heat spreader into Frore's described liquid coldplate architecture, not a demonstrated universal improvement in AI infrastructure economics.
Future qualification will require measurements of processor temperature, liquid conditions, pressure drop and operation under defined workloads. Repeated manufacturing performance and compatibility with specific packages will determine which accelerator designs can use the new coldplate arrangement.
