Aivres Unveils New Liquid-Cooled AI Server Hardware
The KRS8000V4 rack and KL2288 tray enable high-density AI workloads using NVIDIA's new Vera Rubin and B300 series chips.
Updated on Oct. 25, 2026 in Data Centers
Live Poll
Will advancements in AI hardware make large-scale computing more efficient for the average business?
Aivres has unveiled new hardware featuring NVIDIA Vera Rubin CPUs and B300 GPUs at the OCP Global Summit 2026. The launch includes high-density rack systems and compute trays designed for massive AI training and inference workloads.
Why it matters
These systems push the boundaries of liquid-cooled density, allowing data centers to manage extreme memory bandwidth and compute requirements for next-generation AI models. This hardware marks a significant shift toward specialized, high-heat cooling architectures necessary for the latest NVIDIA silicon.
The Aivres KL2288 compute tray delivers 176TB/s of total HBM bandwidth and 2.3TB of HBM4 memory. The larger KRS8000V4 system packs 72 Rubin GPUs and 36 Vera CPUs with over 20TB of aggregate HBM capacity.
The players
Aivres
A hardware manufacturer specializing in high-density server racks and advanced liquid cooling solutions for hyperscale data centers.
NVIDIA
A semiconductor company that designs high-performance GPUs and CPUs, including the Vera Rubin and B300 lines, for artificial intelligence research and commercial deployment.
AMD
A semiconductor firm that produces the EPYC 9006 processors used in conjunction with NVIDIA GPUs in the Aivres compute trays.
The details
The KRS8000V4 system utilizes sixth-generation NVLink, a high-speed interconnect technology that allows GPUs to communicate at lower latency than standard PCIe connections. To maintain thermal stability, the rack employs a 100% liquid-cooled design that operates at 45C. The KL2288 tray further optimizes thermal management with a fanless design, using copper cold plates—conductive metal plates that draw heat away from components—to directly cool the CPUs and DDR5 DIMMs.
Timeline
Aivres showcased the new hardware systems during the 2026 OCP Global Summit.
The Tech Race
The introduction of the KRS8000V4 rack marks a shift in how server manufacturers are addressing the thermal constraints posed by the newest NVIDIA silicon. This development follows the industry-wide trend toward full-rack liquid cooling, directly challenging existing air-cooled enterprise server standards.
Data center operators and large-scale AI developers will be the first to integrate this hardware to accelerate their training and inference workflows. Users should note that these systems require liquid-cooling infrastructure, which may necessitate upgrades to existing facility cooling plants.
The takeaway
The move toward 100% liquid-cooled racks signals that top-tier AI performance now requires dedicated facility-level thermal management. Watch for official data sheet releases and validated thermal efficiency benchmarks from independent third-party laboratories in late 2026.
Further reading
For more on the infrastructure supporting high-density AI workloads, see our coverage of Data Centers.
Source note: This article includes information reported by ServeTheHome.
Live Poll
Will advancements in AI hardware make large-scale computing more efficient for the average business?
