The announced configurations include NVIDIA Vera Rubin systems, BlueField networking hardware and Spectrum-X Ethernet.
Nscale said that, in tests on GB300 NVL72 systems, NVIDIA DSX MaxLPS allowed 37% more GPUs and nearly 50% higher aggregate inference throughput within the same provisioned power budget. It also said it plans Vera Rubin deployments across five European markets. Crusoe is developing liquid-cooled modular AI factories for the Vera Rubin platform, while Nebius plans to use NVIDIA Groq 3 LPX with Vera Rubin NVL72 in its Token Factory.
NVIDIA also said a group of cloud providers, including Nebius, Crusoe, CoreWeave, Lambda and Vultr, support its OpenShell Reference Architecture for constrained agent deployments. The post does not give deployment dates, total new capacity, customer pricing, independently verified performance results, or evidence that these planned systems are generally available.