NVIDIA and CoreWeave launch Vera Rubin NVL72 and Vera CPU for agentic AI workloads
NVIDIA and CoreWeave announced new infrastructure for agentic AI at CoreWeave Fully Connected. The companies introduced the NVIDIA Vera Rubin NVL72 systems with Spectrum-X networking and the NVIDIA Vera CPU, designed for AI agents. Cognition, the lab behind Devin, became the first customer to run production workloads on Vera Rubin, achieving up to a 4.8x increase in token throughput for software…
Key points
- NVIDIA Vera Rubin NVL72 delivers 4.8x higher token throughput for Cognition’s Devin agent in SWE-2 inference
- CoreWeave Forge unifies training, evaluation, and improvement tools with 20% lower failure detection costs
- NVIDIA Vera CPU enables 11,000+ concurrent agent environments with 3x faster sandbox startup times
CoreWeave also launched CoreWeave Forge, a unified environment for training, evaluating, and improving models and agents. Forge integrates tools like Weights & Biases, OpenPipe, and marimo, offering capabilities such as CoreWeave ARIA for experiment analysis, CoreWeave Agent Lens for production observability, and serverless fine-tuning and reinforcement learning. The platform supports open-source models like NVIDIA Nemotron and aims to reduce failure detection costs by 20% while accelerating sandbox startup times by over 3x. Early adopters include healthcare provider Ennoble Care and enterprises like Canva, Capital One, and MasterClass.
The story so far
2 episodes →- NVIDIA and CoreWeave launch Vera Rubin NVL72 and Vera CPU for agentic AI workloadsthis story
From Training to Production, NVIDIA and CoreWeave Close the Loop on Agentic AI
NVIDIA Blog · 30 September 2026
Loading the full article…
This text was published by NVIDIA Blog and written by Stuart Pitts. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- CoreWeave to Offer NVIDIA Vera, the First CPU Built for AI AgentsPress · finance.yahoo.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Cerebras CEO Feldman debates AI scaling limits at Disrupt 2026 · 1 src
- Deepseek releases open-source TileLang for Huawei Ascend chips · 2 src
- Artificial Analysis releases open-source tool to benchmark local AI agents on laptops and workstations · 1 src
- Ollama cheat sheet guides local AI model management · 1 src
- OpenAI VP details Jalapeño ASIC’s AI-assisted design and efficiency focus · 4 src
Comments
via GitHub Discussions