CoreWeave adds HGX B300
CoreWeave added NVIDIA HGX B300 to CoreWeave Cloud and launched Weights & Biases inference capabilities for RL/agent workflows at GTC — and Cline picked CoreWeave for high‑performance autonomous engineering. The platform move and a named customer win arrive alongside multiple ops and supply‑chain job postings this week, signaling a push into larger enterprise deployments. (hpcwire.com, marketscreener.com, linkedin.com)
HGX B300 nodes on CoreWeave ship with about 2.1 TB of HBM3e GPU memory per node — roughly a 50% increase versus the HGX B200 — and the fleet uses liquid cooling for thermal density management. ( ) CoreWeave paired those nodes with NVIDIA’s Quantum‑X800XDR InfiniBand, which the company and press reports say doubles node‑to‑node bandwidth compared with prior links to support low‑latency, high‑bandwidth agent workloads. ( ) CoreWeave closed its acquisition of Weights & Biases on May 5, 2025, a deal that press filings and filings tracked by market outlets pegged at roughly $1.4 billion in transaction value. ( ) The W&B tech that CoreWeave acquired has been embedded into new inference tooling for long‑horizon RL and agent workflows and was demoed during NVIDIA GTC (March 16–19, 2026) as part of the company’s agent‑focused stack. ( ) Cline’s March 17, 2026 announcement names it as an early customer integrating CoreWeave’s W&B Inference into “Cline, The Open and Secure Coding Agent,” positioning the partner to run multi‑step autonomous coding workflows on CoreWeave’s production inference infrastructure. ( ) Hiring activity is extensive: CoreWeave’s careers page and job aggregators show well over 100 open roles across data‑center, operations and supply‑chain functions this month, including a Sr. Manager, Supply Chain Risk, Resilience & Compliance role with a listed Built In San Francisco salary band of approximately $161K–$237K. ( ) As part of the roadmap to scale inference for agentic AI, CoreWeave has signaled plans to add NVIDIA’s Vera Rubin NVL72 platform and Vera CPU racks in the second half of 2026 to complement its Blackwell‑based HGX/GB300 deployments. ( )
