CoreWeave has made a landmark move by becoming one of the first cloud providers to bring NVIDIA’s cutting-edge GB200 NVL72 rack-scale systems online. This launch is giving AI innovators such as Cohere, IBM, and Mistral AI the computing power they need to push the boundaries of next-generation AI models and enterprise applications.

After being the first to make NVIDIA Grace Blackwell generally available, CoreWeave has already achieved record-breaking results in MLPerf benchmarks with the GB200 NVL72 platform. Now, customers gain access to thousands of NVIDIA Blackwell GPUs, transforming CoreWeave’s cloud infrastructure into a state-of-the-art “AI factory” for intelligence at scale.

We work closely with NVIDIA to quickly deliver to customers the latest and most powerful solutions for training AI models and serving inference. With new Grace Blackwell rack-scale systems in hand, many of our customers will be the first to see the benefits and performance of AI innovators operating at scale.

Mike Intrator, CEO of CoreWeave

Cohere: Accelerating Enterprise AI Agents

Cohere is leveraging NVIDIA GB200 NVL72 on CoreWeave’s cloud to develop secure enterprise AI applications. Their enterprise AI platform, North, simplifies building personalized AI agents that automate workflows and surface real-time insights.

By training 100-billion-parameter models on the GB200 NVL72 platform, Cohere has reported up to 3x faster performance compared to previous-gen Hopper GPUs—even before specific Blackwell optimizations. With features like large unified memory, FP4 precision, and the 72-GPU NVLink domain, they are achieving higher throughput and reduced latency for inference.

With access to some of the first NVIDIA GB200 NVL72 systems in the cloud, we are pleased with how easily our workloads port to the NVIDIA Grace Blackwell architecture. This unlocks incredible performance efficiency across our stack — from our vertically integrated North application running on a single Blackwell GPU to scaling training jobs across thousands of them. We’re looking forward to achieving even greater performance with additional optimizations soon.

Autumn Moulder, vice president of engineering at Cohere

IBM: Training Granite Models for Enterprise

IBM is harnessing the power of NVIDIA GB200 NVL72 systems—scaling to thousands of Blackwell GPUs through CoreWeave—to train its next-generation Granite family of open-source, enterprise-ready AI models. These models are optimized for safety, speed, and efficiency, forming the backbone for AI solutions such as IBM watsonx Orchestrate.

IBM’s deployment benefits further from the IBM Storage Scale System, ensuring high-performance storage for AI workloads. IBM clients can access this integrated storage solution within CoreWeave’s dedicated environment.

We are excited to see the acceleration that NVIDIA GB200 NVL72 can bring to training our Granite family of models. This collaboration with CoreWeave will augment IBM’s capabilities to help build advanced, high-performance and cost-efficient models for powering enterprise and agentic AI applications with IBM watsonx.

Sriram Raghavan, vice president of AI at IBM Research

Mistral AI: Scaling Open-Source Language Models

Mistral AI, a leader in open-source language models, is now deploying its first thousand Blackwell GPUs using CoreWeave’s infrastructure. The new resources—backed by NVIDIA Quantum InfiniBand networking—are crucial for the intensive training and deployment cycles needed to advance Mistral’s suite of large language models, like Mistral Large.

Right out of the box and without any further optimizations, we saw a 2x improvement in performance for dense model training. What’s exciting about NVIDIA GB200 NVL72 is the new possibilities it opens up for model development and inference.

Thimothee Lacroix, cofounder and chief technology officer at Mistral AI

Unprecedented Scale for AI Factories

CoreWeave’s deployment allows customers to tap into innovative rack-scale instances with 72 NVIDIA Blackwell GPUs and 36 Grace CPUs per node, and scale to as many as 110,000 GPUs with NVIDIA Quantum-2 InfiniBand networking. This unmatched density and scale enable the rapid development, deployment, and inference of large AI models—fueling both established enterprises and next-gen AI startups.