What happened
CoreWeave announced the production availability of NVIDIA Vera Rubin NVL72 systems and Vera CPUs on its cloud infrastructure. Cognition, the developer of the Devin AI software engineer, is the first customer to run production workloads on this new hardware. Additionally, CoreWeave launched CoreWeave Forge, a unified environment for training and evaluating AI models and agents.
CoreWeave announced the availability of NVIDIA Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet networking on its cloud platform. This makes CoreWeave one of the first cloud providers to deliver this next-generation AI infrastructure to customers. The announcement was made at the CoreWeave Fully Connected event in San Francisco.
Cognition, the applied AI lab behind the Devin AI software engineer, is the first customer to run production workloads on Vera Rubin. Cognition scaled to thousands of GPUs on CoreWeave in nine months and recently benchmarked Vera Rubin’s performance against a GB200 NVL72 baseline using a real-world software engineering workload.
In early tests, Cognition reported that Vera Rubin NVL72 delivered up to a 4.8x increase in total token throughput for SWE-2 workloads compared to GB200 NVL72. These gains translate to faster real-time code generation and more responsive multistep reasoning for Devin.
CoreWeave also announced the availability of NVIDIA Vera, the first CPU built specifically for AI agents. Vera CPUs are designed to handle the high concurrency and isolation requirements of agentic workloads. CoreWeave’s deployment of Vera puts 128 CPUs and 11,264 cores in a single rack, supporting more than 11,000 concurrent environments.
Additionally, CoreWeave launched CoreWeave Forge, a connected environment for training, evaluating, and improving models and agents. Forge unifies Weights & Biases, expertise from OpenPipe, and the open-source marimo notebook project. It is designed to streamline the continuous improvement loop for AI models and agents.
Source details: blogs.nvidia.com ↗
Why it matters
This deployment marks a significant shift in AI infrastructure, moving from generic compute to specialized hardware designed for agentic workloads. The Vera Rubin NVL72 offers substantial performance gains for , while the Vera CPU is optimized for running thousands of isolated agent environments simultaneously. This infrastructure supports the growing demand for low-latency, high-concurrency AI agents that require complex reasoning and , enabling companies to scale agentic applications from prototype to production more efficiently.
The introduction of Vera Rubin NVL72 and Vera CPUs addresses specific bottlenecks in agentic AI, such as the need for low-latency compute and the ability to run thousands of isolated agent environments simultaneously. This specialized infrastructure can significantly reduce the cost and complexity of deploying agentic AI systems.
The 4.8x increase in token throughput reported by Cognition suggests that Vera Rubin NVL72 can handle the high-volume, complex reasoning tasks required by agentic AI more efficiently than previous generations. This could lead to faster development cycles and more responsive AI applications.
CoreWeave Forge aims to solve the fragmentation in AI development tools by providing a unified environment for training, evaluation, and deployment. This can help AI teams iterate more quickly and reduce the signal loss that occurs when moving between different tools and vendors.
The partnership between NVIDIA and CoreWeave highlights the growing importance of co-engineered infrastructure for AI. By working closely with cloud providers, NVIDIA can optimize its hardware and software for specific AI workloads, while CoreWeave can offer differentiated services to its customers.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
In AI, what are a model's "parameters"?
What to watch next
Monitor the adoption of Vera Rubin NVL72 by other major AI labs and enterprises. Watch for further performance benchmarks from Cognition and other early adopters. Observe the expansion of CoreWeave Forge and its integration with other AI tools and frameworks. Track the impact of NVIDIA Vera CPUs on the cost and scalability of agentic AI deployments.
Watch for other major AI labs and enterprises to adopt Vera Rubin NVL72 and Vera CPUs on CoreWeave. The performance and cost benefits reported by Cognition may drive broader adoption.
Monitor further benchmarks and real-world performance data from early adopters of Vera Rubin NVL72 and Vera CPUs. This will provide a clearer picture of the practical benefits of this new infrastructure.
Observe the growth and feature development of CoreWeave Forge. Its ability to integrate with other AI tools and frameworks will be key to its success in the AI development ecosystem.
Track the impact of NVIDIA Vera CPUs on the scalability and cost of agentic AI deployments. The ability to run thousands of isolated environments efficiently could enable new types of agentic AI applications.