CoreWeave First to Run NVIDIA Vera Rubin NVL72; Cognition Signs as Launch Customer
At CoreWeave Fully Connected '26 in San Francisco (4,500+ customers and developers attending), CoreWeave announced it is first to offer NVIDIA Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet networking, and Cognition (the company behind Devin) became the first customer running production workloads on the new infrastructure, with official figures showing roughly 4.8x throughput over GB200. This marks NVIDIA's platform progression from Hopper (2022) and Blackwell (2024) to Vera Rubin, with its first public production deployment.
Confirmed
- CoreWeave announced Vera Rubin NVL72 and Spectrum-X 102.4T Ethernet at Fully Connected '26 (@nordicinst citing the NVIDIA blog, corroborated by @AccBalanced and @ayushthakur0).
- Cognition is the first customer running production workloads on CoreWeave's Vera Rubin, with 4.8x throughput vs GB200 (consistent across posts).
- Analyst Holger Mueller reported that Cognition's Silas Alberti spoke on stage about running on Vera Rubin as CoreWeave's first customer (retweeted by @silasalberti).
- Posts differ slightly on what the 4.8x refers to: @nordicinst cites the NVIDIA blog describing 4.8x training throughput for Devin, while @altryne describes it as token throughput on SWE-2.
Why it matters
- This is the first publicly disclosed production deployment of Vera Rubin NVL72, demonstrating real-world throughput gains of the new NVIDIA generation over GB200 for training- and inference-intensive workloads such as AI coding agents like Devin.
2026-09-30 ~ 2026-10-01 · 6 related posts
Primary sources
- [source] Cognition runs production workloads on Vera Rubin NVL72 with 4.8x token throughput — nordicinst · 2026-09-30
- [source] Cognition Becomes First CoreWeave Vera Rubin NVL72 Customer, Sees 4.8X SWE-2 Throughput Boost — altryne · 2026-10-01
- Cognition Goes Live on Vera Rubin nVL72 at NVIDIA Fully Connected 26 — silasalberti · 2026-10-01
- Cognition First to Run NVIDIA Vera Rubin on CoreWeave, ~4.8x Throughput vs GB200 — silasalberti · 2026-10-01
- CoreWeave first to ship NVIDIA Vera Rubin NVL72, Cognition sees 4.8x throughput vs GB200 — AccBalanced · 2026-10-01
1 near-duplicate retellings: ayushthakur0