Nebius launches serverless AI Cloud 3.5
Nebius rolled out AI Cloud 3.5 with serverless support for NVIDIA RTX PRO 6000 Blackwell GPUs—making prototyping and inference easier without heavy infra ops. That reduces friction for early-stage model testing and could speed hardware/software co-tuning for device teams. (x.com)
Nebius markets the update under the product name "Aether 3.5" and outlines new serverless primitives, cluster-configuration tools and a dedicated Data Transfer Service in its product blog post dated March 26, 2026. (nebius.com) The company says all Aether 3.5 features are live on its global AI Cloud infrastructure, with the new serverless offering entering public preview on March 26, 2026. (markets.financialcontent.com) Aether 3.5 adds the NVIDIA RTX PRO 6000 Blackwell Server Edition to Nebius’s GPU catalog and lists that Server Edition as available on the platform today. (nebius.com) The release highlights a Nebius Data Transfer Service to accelerate model data movement and lists platform-level improvements aimed at simplifying cluster configuration and streamlining data operations. (nebius.com) Nebius is scaling physical capacity in parallel: management has stated a near-term objective of roughly 1 GW of power with a pipeline beyond 3 GW, and the company secured local approval for an Independence, Missouri AI factory with up to 1.2 GW potential. (computing.net) Wall Street attention followed the launch: Bank of America initiated coverage with a Buy rating and a $150 price target this week, implying roughly 30% upside from recent levels. (marketbeat.com) Competing clouds are moving similarly: Google Cloud announced Cloud Run support for NVIDIA RTX PRO 6000 Blackwell Server Edition in a serverless preview, signaling parallel vendor moves to offer Blackwell-class GPUs in serverless inference flows. (cloud.google.com)
