CoreWeave announced new infrastructure additions to its AI-native cloud platform centered on general availability of NVIDIA HGX B300, part of the Blackwell Ultra platform, for production-scale AI workloads. The company said the offering is designed to support frontier model development, reinforcement learning, and agentic AI workloads, with each node configured with up to 2.1 TB of HBM3e memory, NVIDIA Quantum-X800XDR InfiniBand, and liquid cooling.
CoreWeave is also positioning the launch as a shift toward infrastructure built not only for training but for inference and agentic workloads that require large context windows, lower latency, and more persistent orchestration. The company said its platform also integrates Weights & Biases tools, including W&B Inference and Weave, to help customers move models into production more efficiently.
The release is notable because it reflects where demand appears to be heading across the AI cloud market. AI infrastructure providers are increasingly differentiating themselves not simply on raw GPU availability, but on how well their platforms support production-scale inference, observability, and complex multi-step AI applications. CoreWeave also said it expects production availability of NVIDIA Vera Rubin NVL72 and Vera CPU racks in the second half of 2026.
Key points
• CoreWeave announced general availability of NVIDIA HGX B300 on its cloud
• Positioned for frontier AI, reinforcement learning, and agentic workloads
• Configurations include up to 2.1 TB of HBM3e memory per node
• Includes NVIDIA Quantum-X800XDR InfiniBand and liquid cooling
• CoreWeave also pointed to Vera Rubin NVL72 production availability in 2H 2026
“The next phase of AI requires infrastructure purpose-built not only for training, but for production-scale inference and agentic systems,” said CoreWeave in its announcement. “By bringing NVIDIA HGX B300 to general availability and integrating observability and developer tools, we are helping customers move faster from experimentation to production.”
🌐 Analysis
This is an important cloud infrastructure story because it shows how the market narrative is shifting from model training to production-scale AI operations. CoreWeave is clearly trying to position itself for the next demand wave: reasoning, inference, and agentic applications that need large memory footprints and tightly integrated operational tooling.
