NVIDIA cloud partners expand regional compute, model services and agent controlsMachine translation

Nscale reports that its test setup supported 37% more GPUs and nearly 50% higher overall inference throughput under the same preconfigured power budget; Crusoe is developing a liquid-cooled AI factory for Vera Rubin. Nebius launched a robotics AI cloud offering, while Baseten built a code-review routing model for CodeRabbit. Fireworks and Era4 are collaborating on enterprise AI plans, and Mistral and Verda plan to deploy Vera Rubin infrastructure. NVIDIA announced the OpenShell reference architecture to help organizations limit what AI agents can access and do.