CoreWeave and NVIDIA: From Training to Production with AI

CoreWeave has collaborated with NVIDIA for nearly a decade to create a cloud infrastructure specifically designed for AI. The partnership results in a platform available for deploying the latest NVIDIA Vera Rubin NVL72 systems and advanced networking solutions. During the CoreWeave Fully Connected event in San Francisco, it was announced that Cognition, the AI lab behind Devin, is the first customer to utilize these systems for operational workloads.

The new NVIDIA Vera CPU, aimed at agentic AI workloads, enhances CoreWeave's capacity with capabilities for thousands of simultaneous agent-environments. This infrastructure utilizes powerful technologies such as Spectrum-X Ethernet networks to ensure low latency and high efficiency. Cognition reports a significant increase of 4.8x in token throughput for software engineering workloads during early tests on Vera Rubin, leading to faster and more efficient code generation.

CoreWeave also introduces CoreWeave Forge, an integrated environment that promotes the training, evaluation, and improvement of AI models and agents. This platform utilizes various tools and offers functionalities such as CoreWeave ARIA and Agent Lens for continuous improvement of models and agents. The innovations serve not only startups but also large enterprises such as Canva and Capital One, which are accelerating their AI projects into production.

With record results in machine learning tests, CoreWeave has confirmed its position as a leading cloud provider for AI. The collaboration with NVIDIA enables clients to convert experimental AI agent systems into operational systems that add real value in sectors like healthcare, where CoreWeave is deployed for clinical AI inference and automation of process documentation.

Read the full article from Nvidia.