CoreWeave announced the production availability of NVIDIA Vera Rubin NVL72 systems featuring Spectrum-X Ethernet networking alongside the launch of CoreWeave Forge, an environment for training and evaluating AI agents. The cloud provider is among the first to deliver the Vera Rubin platform to customers, expanding its long-standing infrastructure partnership with NVIDIA. CoreWeave will also offer NVIDIA’s Vera CPU, which is specifically designed for agentic AI workloads.
Cognition, creator of the Devin AI software engineer, has deployed production workloads on the Vera Rubin infrastructure. In early benchmark tests using software engineering tasks from FrontierCode, Cognition recorded up to a 4.8x increase in total token throughput for SWE-2 inference workloads compared to a GB200 NVL72 baseline.
These throughput gains enable faster real-time code generation and multi-step reasoning for agentic workloads. CoreWeave is offering access to the new hardware through its suite of management and inference services, addressing the high-concurrency and low-latency demands of agentic AI deployments.
Why it matters
Up to 4.8x throughput improvements on software engineering tasks lower token costs and enable faster multi-step agent reasoning.
Purpose-built CPUs and infrastructure for agents signal a shift toward hardware optimized for high-concurrency, long-context workloads.
Early cloud access to next-gen compute offers a competitive performance advantage for resource-intensive AI operations.
Source: blogs.nvidia.com



