Nexalus and Renda have launched the Renda Nova8, a liquid-cooled AI server designed to retrofit into standard 19-inch racks in existing air-cooled data centres. The 4U server integrates eight Nvidia Blackwell GPUs, CPUs, memory, storage, power, and Nexalus’s proprietary jet cooling technology—eliminating the need for new piping or cooling infrastructure. According to the companies, it delivers up to 22% better energy efficiency and 20% higher AI token throughput compared to equivalent air-cooled systems.
Bridging the cooling gap in legacy facilities
Many data centres face physical and financial constraints when upgrading for high-density AI workloads. The Nova8 is engineered to bypass these barriers by fitting into existing rack footprints without requiring facility-level liquid cooling retrofits. Nexalus’s jet cooling system directs coolant onto cold plates rather than relying on channel-based flow, reportedly enabling full thermal management of all components within a single chassis. This approach draws from thermal lessons in Formula 1, where Renda and Nexalus work with eight of the 11 Formula 1 constructors—applying extreme-performance cooling logic to data centre scale.
Power and infrastructure challenges persist
While the Nova8 addresses cooling, broader power infrastructure issues still limit AI deployment. A Fluke survey of 500 US data centre operators found that 74% face moderate to severe limitations due to insufficient power or grid capacity. Only 36% said their electrical systems are fully ready for high-density AI workloads. Power-quality issues—including voltage sags, harmonics, and thermal hotspots—have already impacted deployment or operations for nearly a third of respondents. These problems often arise at the intersection of power, cooling, and cabling, where undocumented or degraded infrastructure undermines reliability.
High-voltage demands raise commissioning risks
As AI facilities grow, many are opting for direct high-voltage (HV) grid connections—66kV, 110kV, or higher—to meet multi-megawatt demands. Such connections introduce complex HV infrastructure—substations, power transformers, protection relays—into the data centre’s critical path. Commissioning these systems requires specialist expertise distinct from low-voltage practices, with risks including arc flash, equipment damage, and grid disruption. Experts stress that HV commissioning must be integrated early in project planning, with rigorous testing of transformers, cables, switchgear, and protection systems before energization.
What it means
The Nova8 offers a pragmatic upgrade path for operators in the UK, UAE, US, and Kazakhstan who lack the time or capital for full facility retrofits. However, improved cooling alone won’t overcome power-quality or grid-access constraints. Organisations should assess not just thermal capacity but also electrical integrity, monitoring coverage, and workforce readiness. Real-time visibility is lacking in 78% of facilities, and 15% only detect electrical issues after failures occur. For AI scalability, cooling innovation must be matched with infrastructure diagnostics and specialist commissioning.
Organisations considering retrofit solutions should verify local power quality, ensure testing occurs at load points, and involve HV specialists early if scaling beyond 10MW. The Nova8 reduces one dependency—liquid cooling infrastructure—but does not eliminate systemic electrical risks.