Vendors & EquipmentNova8Liquid CoolingAI ServersNvidia Blackwell

Nexalus, Renda launch Nova8 eight-GPU liquid-cooled 4U rack

Renda and Nexalus launched the Nova8, an eight-GPU, 4U liquid-cooled server designed for standard 19-inch racks in existing air-cooled data centers without new pipework.

3 min read

Nexalus and Renda launch liquid-cooled AI server for data center retrofit marketVendors & Equipment
Nexalus and Renda launch liquid-cooled AI server for data center retrofit marketann_zima / pixabay

Why it matters

  • The Nova8 combines eight Nvidia Blackwell GPUs and integrated server components in one 4U chassis.
  • The server fits standard 19-inch racks without new piping or cooling infrastructure, according to the vendors.
  • Nexalus claims 22% better energy efficiency and up to 20% more AI token throughput than a similar air-cooled setup.
  • The companies’ Formula 1 cooling work includes eight of the championship’s 11 constructors.

The story

By James Calloway

Nexalus and Renda launched the Nova8, a liquid-cooled 4U server built around eight Nvidia Blackwell GPUs that fits standard 19-inch racks in existing air-cooled data centers.

The launch addresses a facility constraint rather than a new-build cycle. In a joint statement, the companies said operators can swap Nova8 for current equipment without adding new piping or cooling infrastructure. The design therefore targets operators that need more throughput but cannot readily retrofit their facilities for liquid cooling.

That distinction defines the product’s commercial pitch: rack-based liquid cooling for facilities that cannot readily retrofit for it. The announcement did not identify a customer, order, price, shipping date or independent performance test. Immediate uptake remains unstated. Independent proof remains absent.

  • Compute: Eight Nvidia Blackwell GPUs, plus integrated CPUs, memory, storage, power and cooling in one server.
  • Format: A single 4U chassis, mounted in a standard 19-inch rack.
  • Cooling: Nexalus jet cooling, which squirts coolant onto the liquid-cooling cold plate instead of routing it through channels.

Nexalus says the jet-cooling arrangement can cool the CPU and all eight GPUs at full power within the 4U enclosure. Against a similar air-cooled configuration, the company claims 22% better energy efficiency and up to 20% more AI token throughput. Those figures come from Nexalus. The release provides no test protocol or third-party validation.

The comparison is specific: the efficiency figure is stated against a similar air-cooled setup, while the throughput figure concerns AI token production. The announcement gives no workload, model size or absolute power consumption, preventing readers from translating either percentage into a bill or output figure.

Kenneth O’Mahony, Nexalus CEO, said: “Operators need more throughput, but most cannot pause to retrofit their facilities for liquid cooling. Our patented, proprietary server design removes that barrier, bringing liquid-cooled performance into the racks they already run.”

The design drew on the companies’ experience cooling GPU servers in Formula 1 racing. They said that motorsport work informed Nova8’s development, and their work covers eight of the 11 constructors in the championship. The commercial product remains a rack-mounted data-center server; Formula 1 supplied the cooling experience.

Renda’s Ian “8Pack” Parry said: “Formula 1 taught us that thermal management decides how much performance you can actually use, and AI is now facing the same challenge at data center scale.”

For operators, the citation establishes the vendors’ thermal pedigree, but it does not substitute for workload-specific server benchmarks.

Nexalus is headquartered in Cork, Ireland. It already works with server makers including HPE, which uses Nexalus’s liquid-cooling technology in its systems. Renda, the server maker, is based in the UK.

What changes for data-center operators?

For operators already using standard 19-inch racks in air-cooled facilities, the immediate change is deployment format. Nova8 combines eight Blackwell GPUs, the server platform and liquid cooling in one 4U chassis, without requiring new cooling pipework at the facility.

The practical benefit is narrower facility work, according to both vendors. Operators do not have to commission a separate liquid-cooling infrastructure project before installing the system. That matters for sites seeking more throughput while retaining an existing facility, although the release does not provide a named installation or delivery schedule.

The 22% energy-efficiency gain and 20% token-throughput increase remain the main commercial arguments, but they are supplier claims rather than demonstrated customer results. No buyer or operating data center was named, so the claims cannot yet be separated from vendor marketing. Pricing, purchase volumes and deployment dates also remain absent.

The retrofit proposition gives Nova8 a route into established data centers that have not planned for liquid cooling. Its wider adoption still depends on commercial deliveries, operator acceptance and independent performance evidence. Existing operators can assess it as replacement equipment, but broad deployment has not been demonstrated.

The next commercial test will come from named operator deployments and disclosed delivery schedules, which the launch did not provide. Until those details and independent operating results emerge, the scale and timing of Nova8’s retrofit-led rollout remain unproven.

Also reported

Share this article:

« Previous

More from James Calloway

James Calloway

Show full bio

Staff writer covering companies, deals and strategy in Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.

565 articles