At a glance
| Booth | I3000 |
| Country | GB |
| Website | www.infiniflux.cool |
Company profile
INFINIFLUX is a deep-tech startup developing advanced two-phase Direct-to-Die liquid-cooling technologies for high-power 2.5D and 3D chips used in AI and HPC data centres. Its technology targets hotspots directly on processor dies to remove heat where it is concentrated, with the stated aims of eliminating heat spikes, increasing compute capacity and extending device lifespan. The cooling concept is based on a patent-pending cold-plate design with 75 percent lower thermal resistance and a system architecture requiring substantially less pump energy. INFINIFLUX states that its two-phase system removes more heat than single-phase liquid cooling, uses less energy and requires virtually no maintenance. The technology is intended for both new-build data centres and retrofit projects, allowing the same cooling approach to be used in new infrastructure and existing facilities. The company’s mission is to make AI infrastructure Net Zero by reducing the energy and water burden associated with high-density compute cooling. INFINIFLUX describes deep roots in the United Kingdom scientific community and links its technology development to research experience in advanced and cryogenic cooling. One founder’s academic work focused on cryogenic cooling systems for quantum technologies near absolute zero at -272 degrees Celsius and included a silicon-based cooler that culminated in a patent filing. The company positions this cooling background as relevant to removing thermal bottlenecks in present high-power computing and to future 3D and quantum-chip designs. INFINIFLUX has secured equity investment from Zinc VC and grant funding from Innovate UK, the Henry Royce Institute and the Grantham Institutes. It has won the IC Taiwan Grand Challenge award, showcased its work at Innovex and Computex, and reported interest from semiconductor and compute-ecosystem companies in Taiwan and the wider Asian region.
Exhibits
ChipCool is INFINIFLUX’s two-phase Direct Liquid Cooling approach for high-power GPUs and processors, designed to keep chips within temperature limits as Thermal Design Power and rack density increase. The company states that ChipCool can keep chips 10 to 15 degrees Celsius cooler, reduce throttling, increase IT workload capacity and prolong chip life. Its two-phase refrigerant system is described as PFAS-free with zero Global Warming Potential and zero Ozone Depleting Potential, while passive flow and water-free refrigerant are used to target zero-maintenance operation. INFINIFLUX reports a PUE below 1.03 and uses Performance Per Watt, measured in TFLOPS per watt, as a key metric alongside conventional data-centre efficiency measures. The ChipCool cold plate is described as mechanically robust and sufficiently thin to reach an ultra-low thermal resistance of 0.009 K/W. Its system is stated to require between 5 and 20 times less pump power than conventional direct-liquid-cooling systems. For data-centre operators, listed benefits include less throttling, higher rack density, a 90 percent cooling-energy reduction compared with air cooling, near-zero water-usage effectiveness, retrofit to the facility water loop and heat-reuse potential. For server OEMs, the company lists the ability to raise GPU TDP, reduce chassis volume by up to 25 percent and extend component life by 2X, while GPU producers are offered additional thermal headroom, a 10-degree chip-temperature drop and support for higher performance or smaller die. ChipCool uses a hermetically closed server-level Third Loop in which refrigerant does not contact other liquids in the system, preventing cross-contamination and removing water-related bio-fouling risk from that loop. The architecture is designed to work with primary loops and cooling-distribution units from multiple manufacturers, reducing vendor lock-in and supporting retrofit into existing liquid-cooling installations. INFINIFLUX states that the pressure drop the CDU must overcome is about 80 percent lower than in conventional direct liquid cooling, providing additional pump-energy savings. The technology also targets future 3D-stacked chips, where trapped hotspots and steep temperature gradients create thermo-mechanical stress, aging, thermal runaway and runtime throttling. The company is developing cooling intended to make high-power 3D AI stacks practical by removing the vertical heat constraint that otherwise limits dense logic-and-memory integration. Its collaboration model includes joint feasibility studies with advanced-packaging companies to integrate cold-lid technology into packages and partnerships with MEMS foundries for silicon cold-lid production. Joint studies with server OEMs and ODMs focus on integrating the two-phase cooling loop into servers, while component manufacturers, rack manufacturers and system integrators can contribute to the wider cooling system. Pilot projects with data centres using advanced liquid cooling and collaborations with academic groups in liquid cooling, thermo-hydraulics and IC packaging provide routes for evaluating the technology at package, server and facility levels. The company describes the rapid rise of AI compute as pushing single-phase copper cold-plate cooling toward its heat-flux and pump-power limits, which is the operating context for its two-phase Direct-to-Die approach. ChipCool is therefore positioned for both immediate retrofit of existing liquid-cooled infrastructure and future high-density AI and HPC systems with advanced packages and 3D chip architectures.
Capabilities and products
- R&D of PFAS-free two-phase direct-to-die liquid cooling technology
- cold-plate design with ultra-low thermal resistance of 0.009 K/W
- hermetically sealed server-level refrigerant loops compatible with existing CDUs