CoreWeave became one of the first cloud providers offering NVIDIA GB200 NVL72 systems at scale. Companies like Cohere, IBM, and Mistral AI already use these systems to train advanced AI models. The company made NVIDIA Grace Blackwell generally available after showing strong results in MLPerf benchmarks. Customers can access thousands of these powerful GPUs designed for reasoning and AI agents.
Mike Intrator, CoreWeave CEO, said they work closely with NVIDIA to deliver the latest solutions for AI model training. The systems transform data centers into AI factories that create intelligence at scale. Leading AI companies worldwide have started using GB200 NVL72 for various applications, including agentic AI and cutting-edge model development.
Cohere reports up to three times better performance when training 100 billion-parameter models compared to previous NVIDIA Hopper GPUs. IBM uses these systems to train next-generation Granite models, which provide enterprise-ready AI with safety features. Mistral AI received its first thousand Blackwell GPUs to develop open-source language models.
The Paris-based company saw double the performance for dense model training without additional optimizations. CoreWeave offers instances with rack-scale NVIDIA NVLink across 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs. These instances scale up to 110,000 GPUs with NVIDIA Quantum-2 InfiniBand networking.
Mike Intrator, CoreWeave CEO, said they work closely with NVIDIA to deliver the latest solutions for AI model training. The systems transform data centers into AI factories that create intelligence at scale. Leading AI companies worldwide have started using GB200 NVL72 for various applications, including agentic AI and cutting-edge model development.
Cohere reports up to three times better performance when training 100 billion-parameter models compared to previous NVIDIA Hopper GPUs. IBM uses these systems to train next-generation Granite models, which provide enterprise-ready AI with safety features. Mistral AI received its first thousand Blackwell GPUs to develop open-source language models.
The Paris-based company saw double the performance for dense model training without additional optimizations. CoreWeave offers instances with rack-scale NVIDIA NVLink across 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs. These instances scale up to 110,000 GPUs with NVIDIA Quantum-2 InfiniBand networking.
