CoreWeave shows production‑scale demos at GTC
CoreWeave showcased production‑scale AI at GTC, demoing observability, liquid‑cooled hardware, and end‑to‑end MLOps with partners — a concrete bridge from lab prototypes to deployable stacks. The booth demos underscored how hardware, cooling, and ops tooling are converging for real‑world deployments. (x.com)
CoreWeave announced general availability of NVIDIA HGX B300 on its cloud, with nodes featuring 2.1 TB of HBM3e memory and next‑gen NVIDIA Quantum‑X800XDR InfiniBand for higher node‑to‑node bandwidth. (morningstar.com) CoreWeave ran an in‑person GTC session (S82200) that demoed CoreWeave Mission Control and listed Corey Sanders, Chen Goldberg, and Sebastian Yunge as presenters in the March 16 session. (nvidia.com) Mission Control’s expanded capabilities—announced in December—include Telemetry Relay for streaming audit logs to customer SIEMs, GPU Straggler Detection for rank‑level diagnosis, and a Mission Control Agent currently in preview. (secure.businesswire.com) CoreWeave highlighted tighter integrations with Weights & Biases at GTC, rolling out Serverless RL, W&B Weave, and a W&B mobile app (iOS), while Serverless RL was touted to deliver ~1.4× faster training, up to 5× lower costs and 60× lower latency for agent workflows. (coreweave.com) The company is the first hyperscaler to deploy NVIDIA GB300 NVL72 in Dell integrated racks—each rack packs 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs and expands HBM3e memory to as much as ~21 TB per rack. (storagereview.com) Those GB300 NVL72 racks are liquid‑cooled deployments that draw roughly 140 kW at full load, underscoring the cooling and power engineering CoreWeave highlighted on the show floor. (vertiv.com) CoreWeave also promoted new consumption options such as Flex Reservations for guaranteed peak capacity and referenced a $5 billion revenue scale for 2025 as part of its growth narrative heading into 2026. (constellationr.com)
