CoreWeave puts Nvidia Vera Rubin NVL72 into production with Cognition

At Fully Connected in San Francisco on September 30, CoreWeave said Nvidia Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet are now available on its cloud, with Cognition the first customer running production Devin workloads. Nvidia's blog quoted HPC vice president Ian Buck and said CoreWeave received its first production racks earlier in September, then stood up a Cognition cluster in days. Cognition reported up to 4.8 times total token throughput on SWE-2 inference versus a GB200 NVL72 baseline, and CoreWeave separately cited a 3.8 times output-token gain on reinforcement-learning jobs; those figures are customer benchmarks, not Nvidia specifications. CoreWeave is untracked here.