Vera Rubin NVL72 Racks Push 227 kW Power, NVLink 6 Fabric and Liquid Cooling
Infrastructure buyers must plan for 227 kW rack power, mandatory liquid cooling, NVLink 6 copper spine and HBM4 supply risk in Vera Rubin NVL72.
Read more
Infrastructure buyers must plan for 227 kW rack power, mandatory liquid cooling, NVLink 6 copper spine and HBM4 supply risk in Vera Rubin NVL72.
Read more
Engineering comparison of cloud computing AI chips from AWS (Trainium/Inferentia), Google (TPU v6e & Ironwood) and Microsoft (Azure Maia) in hyperscaler data centres.
Read more
The 400G-to-800G-to-1.6T OSFP transition brings 500m DR4 and 500km coherent reach, 36 ports per 1U, and new DSP thermal budgets for AI data centres.
Read more
Data centre operators evaluating GB300 to Vera Rubin racks need to plan for 142 kW power envelopes, liquid cooling with leak detection, and NVLink domain-aware scheduling before 2027 volumes.
Read moreOur Data Centres hub delivers hardware engineering coverage on AI chips data center news today and cloud computing AI chips. As GPU and custom XPU thermal design power (TDP) climbs past 1,200 W to 2,300 W per socket, the unit of compute has shifted from the individual server chassis to the liquid-cooled rack pod.
Moving from 142 kW GB300 NVL72 cabinets to 227 kW Vera Rubin NVL72 racks requires warm-water single-phase cold plates, redundant Coolant Distribution Units (CDUs), 48V/54V DC busbars and 1.6T OSFP optical transceivers engineered for tight thermal budgets.
Hyperscale cloud providers are deploying custom accelerators in dedicated liquid-cooled pods. Explore our in-depth guide to AWS (Amazon) Trainium2/Trainium3 UltraServers, Google Cloud TPU v6e/Ironwood 9,216-chip pods and Microsoft Azure Maia, plus our Buyer’s Guide to AI server power and cooling.