TL;DR
IT Times reported on October 26 that AI infrastructure vendors showcased liquid-cooled compute hardware based on NVIDIA's new Vera Rubin platform at the OCP Global Summit in San Jose. The Vera Rubin NVL72 liquid-cooled AI rack integrates 72 Rubin GPUs and 36 Vera CPUs with over 20 TB of onboard HBM, running 100% liquid-cooled at a 45 degrees Celsius water temperature.
Event Overview
On October 26, 2026, IT Times reported that at the OCP (Open Compute Project) Global Summit in San Jose, AI infrastructure vendors exhibited liquid-cooled compute hardware based on NVIDIA's next-generation Vera Rubin platform, including the Vera Rubin NVL72 liquid-cooled AI rack, the HGX Rubin NVL8 compute tray and HGX B300 servers.
Key Information
| Item | Data |
|---|---|
| NVL72 compute trays | 18 x 1U compute trays |
| NVL72 NVLink switch trays | 9 x 1U NVLink switch trays |
| Integrated GPUs | 72 Rubin GPUs |
| Integrated CPUs | 36 Vera CPUs |
| Onboard HBM | Over 20 TB |
| Cooling | 100% liquid-cooled, running at 45 degrees Celsius |
Hardware on Display
- Vera Rubin NVL72 liquid-cooled AI rack: Uses 18 x 1U compute trays and 9 x 1U NVLink switch trays, integrating 72 Rubin GPUs and 36 Vera CPUs over sixth-generation NVLink, with over 20 TB of onboard HBM. The rack is 100% liquid-cooled and can run at a water temperature of 45 degrees Celsius.
- 2U liquid-cooled HGX Rubin NVL8 compute tray: Equipped with 8 Rubin GPUs and dual-socket x86 processors, delivering up to 400 PFLOPS inference at NVFP4 precision and 280 PFLOPS training, with 2.3 TB of HBM4 memory and 176 TB/s total bandwidth.
- HGX B300 servers: Also exhibited on site.
Background and Impact
High-density liquid-cooled racks target LLM training and inference scenarios and represent the compute form of next-generation AI data centers. As per-rack GPU density keeps rising (the NVL72 packs 72 GPUs per rack), liquid cooling and 45 degrees Celsius hot-water operation become key levers for cutting data center cooling energy. The OCP summit is a bellwether for open compute infrastructure; the concentrated debut of Vera Rubin hardware provides a reference for next-generation AI data center deployment in 2026-2027.
Official Sources
- IT Times: NVIDIA Vera Rubin liquid-cooled AI racks debut at OCP Summit (October 26, 2026) link