CoreWeave Inc. (NASDAQ: CRWV) announced the general availability of NVIDIA HGX B300 systems on its cloud platform at NVIDIA's GTC conference. The company also introduced new capabilities developed with Weights & Biases for reinforcement learning and agent development workflows.

The NVIDIA HGX B300, part of the Blackwell Ultra platform, features 2.1 TB of HBM3e memory, representing a 50% increase over HGX B200 instances. The systems support NVIDIA Quantum-X800XDR InfiniBand connectivity and use liquid cooling technology.

"The next phase of AI is being defined by how efficiently AI systems can run and scale in production," said Michael Intrator, CEO and co-founder of CoreWeave. "By pairing the massive compute power of NVIDIA's latest hardware with CoreWeave's cloud services, we're enabling enterprises to build and refine autonomous agents faster and more reliably than ever before."

CoreWeave expects to deploy NVIDIA Vera Rubin NVL72 platforms and NVIDIA Vera CPU racks in the second half of 2026. The company also introduced Weights & Biases Serverless RL, which eliminates the need for simulated environments in reinforcement learning training.

The platform expansion includes production agent evaluations through W&B Weave and W&B Models for robotics applications. CoreWeave also released the first iOS mobile app for Weights & Biases, allowing researchers to monitor AI model training remotely.

The company has achieved top MLPerf benchmarks and received NVIDIA Exemplar Cloud validation for both training and inference on GB200 NVL72 systems. CoreWeave completed its public listing on Nasdaq in March 2025.