Overview
- CoreWeave on Wednesday operationalized seven Nvidia Vera Rubin NVL72 racks across two regions, bringing 504 Rubin GPUs into multi‑rack production as it moves beyond its single‑rack proof of concept.
- The company launched CoreWeave AI Object Storage features including LOTA, which it says can cut storage read latency up to eight times and deliver as much as 7 GB/s per GPU, alongside Racky for unified rack control and Valvey for per‑rack liquid cooling management.
- CoreWeave disclosed a private placement of $3.0 billion of convertible senior notes with an initial purchaser option for $500 million and opened an at‑the‑market program for up to 35 million shares, and it plans capped calls to limit dilution.
- Investors reacted with short‑term volatility as analysts and outlets flagged strong demand—a Q2 backlog near $104.2 billion and contracted power rising toward about 4.2 GW—alongside heavy liabilities, rising interest expense, and very large planned capex that make timely execution critical.
- Management expects broader Rubin cluster availability later in 2026 and potential hyperscaler adoption, a development that could speed large‑model training but depends on CoreWeave scaling networking, power, cooling and storage reliably at much larger footprint.