What happened
Nexalus and RENDA announced the Nova8, a liquid‑cooled AI server designed to fit into existing 19‑inch, air‑cooled data‑center racks. The 4U chassis houses eight NVIDIA RTX PRO 6000 Blackwell GPUs, each capable of 600 W, and uses Nexalus’s jet‑impingement cooling to remove heat without requiring building‑level liquid piping. The company claims up to 20% higher AI throughput and 22% better energy efficiency versus comparable air‑cooled systems, citing internal Llama 3.3 tests that achieved 55,000 tokens per second at 8.6 tokens per watt. The launch includes a “Test Drive” program that lets qualified prospects remotely evaluate the hardware, with after‑sales support from UNICOM Engineering.
On November 17, Nexalus and RENDA released a joint announcement describing the Nova8 server, positioning it as a retrofit‑friendly solution for data‑centers that currently rely on air cooling. The server’s 4U form factor conforms to standard 19‑inch rack dimensions, allowing it to be installed without structural modifications.
The hardware integrates eight NVIDIA RTX PRO 6000 Blackwell GPUs, each rated for a full 600 W power draw. Cooling is achieved through Nexalus’s jet‑impingement technology, which directs high‑velocity coolant jets at a plate above the hottest chip regions, rather than relying on passive channel flow. The system includes ten redundant pumps and sixteen fans to maintain thermal stability.
In the launch materials, Nexalus claims the Nova8 can deliver up to 20% higher AI throughput and 22% better energy efficiency compared with comparable air‑cooled servers. Internal testing on a Llama 3.3 model reportedly reached 55,000 tokens per second with an efficiency of 8.6 tokens per watt. The company emphasizes that these numbers are not independently verified and may vary with different workloads, levels, and system configurations.
To facilitate evaluation, the vendors are offering a remote “Test Drive” program for qualified prospects, allowing potential customers to run their own workloads on the hardware before committing to purchase. Global after‑sales support is provided through UNICOM Engineering, which will handle installation, maintenance, and troubleshooting.
Why it matters
The Nova8 targets a growing pain point for enterprises that need more AI but lack the budget or time to overhaul cooling infrastructure. By delivering a self‑contained liquid‑cooling loop that fits standard racks, the server could enable faster GPU deployment in legacy facilities, potentially reducing capital expenditures and accelerating AI workloads. If the claimed efficiency gains hold in real‑world settings, operators could lower electricity costs and improve rack density, which is critical as AI models grow larger and demand more power. However, the performance and efficiency figures are company‑provided and not independently verified, so buyers will need to validate the hardware against their specific models, settings, and workload characteristics. The product also raises operational considerations around maintenance, fault detection, and serviceability that differ from traditional air‑cooled servers.
The AI market is increasingly constrained by data‑center power and cooling limits. Traditional air‑cooled racks often hit thermal ceilings before GPU capacity is fully utilized, forcing organizations to either under‑utilize expensive hardware or invest in costly cooling upgrades. A self‑contained liquid‑cooled server like the Nova8 promises to bypass these constraints, potentially unlocking higher GPU utilization without major facility retrofits.
Energy efficiency is a critical cost driver for large‑scale AI deployments. If the Nova8’s claimed 22% efficiency improvement holds in practice, operators could see measurable reductions in electricity bills and carbon footprints, aligning with both financial and sustainability goals.
The product’s compatibility with existing rack infrastructure lowers the barrier to entry for mid‑size enterprises and research labs that lack the capital to build dedicated liquid‑cooled aisles. This could broaden the market for high‑performance AI hardware beyond hyperscale cloud providers.
However, the lack of third‑party validation means the real‑world impact remains uncertain. Performance can vary widely based on model architecture, batch size, (FP16 vs. BF16), and software stack. Buyers will need to conduct thorough testing to confirm that the advertised gains translate to their specific use cases.
Operationally, introducing liquid cooling adds new maintenance considerations—pump reliability, coolant leakage risk, and component replacement procedures—that differ from familiar air‑cooled server practices. Organizations must assess whether their facilities teams have the expertise or support contracts to manage these aspects without compromising uptime.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Key indicators to monitor include independent results, customer case studies confirming the claimed throughput and efficiency gains, and the adoption rate of the Test Drive program. Observers should also watch for any reported issues with the self‑contained cooling loop, such as pump reliability or maintenance complexity, as well as how data‑center operators integrate the Nova8 into existing power and networking budgets.
Independent publications (e.g., MLPerf, TechPowerUp) publishing head‑to‑head comparisons of Nova8 versus leading air‑cooled servers.
Customer case studies or press releases confirming real‑world throughput and power savings, especially from enterprises with diverse AI workloads.
Reports on the reliability of the jet‑impingement cooling system, including pump failure rates, coolant maintenance intervals, and any incidents of leakage or overheating.
Adoption metrics for the Test Drive program, such as the number of qualified prospects, conversion rates, and feedback from early users regarding integration effort and performance.
Potential regulatory or compliance considerations related to liquid cooling in data‑centers, especially in regions with strict environmental or safety standards.