NVIDIA Grace Blackwell GPUs Powering the Next Wave of AI Innovation at CoreWeave

Introduction to the NVIDIA Grace Blackwell Architecture

CoreWeave has announced a significant milestone in the advancement of artificial intelligence, becoming one of the first cloud providers to deploy NVIDIA GB200 NVL72 systems at scale. This deployment brings thousands of NVIDIA Blackwell GPUs online, poised to revolutionize the development and deployment of next-generation AI models and applications. AI frontier companies, including Cohere, IBM, and Mistral AI, are already leveraging this powerful infrastructure to push the boundaries of what's possible in AI. The NVIDIA GB200 NVL72 represents a leap forward in accelerated computing, designed specifically for the demanding requirements of AI inference and agentic workloads. This initiative transforms cloud data centers into highly efficient "AI factories," capable of processing vast amounts of data and converting it into actionable intelligence with unprecedented speed and accuracy.

CoreWeave's Commitment to Cutting-Edge AI Infrastructure

Mike Intrator, CEO of CoreWeave, emphasized the company's close collaboration with NVIDIA to deliver state-of-the-art solutions for AI training and 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," Intrator stated. This sentiment underscores CoreWeave's dedication to providing its customers with access to the most advanced and powerful computing resources available. The successful integration and scaling of these NVIDIA Blackwell GPUs demonstrate CoreWeave's engineering prowess and its commitment to enabling the future of AI development. The company's proactive approach ensures that its clients are at the forefront of AI innovation, equipped with the tools necessary to tackle complex challenges and achieve groundbreaking results.

Accelerating Personalized AI Agents with Cohere

Cohere, a leader in AI development, is utilizing the NVIDIA GB200 NVL72 systems on CoreWeave to enhance its capabilities in creating personalized AI agents. Cohere's North platform, designed for enterprises, enables the development of secure, personalized AI agents that can automate workflows and provide real-time insights. Early results from Cohere's deployment have shown remarkable performance gains. Specifically, the company is experiencing up to a 3x increase in training performance for models with 100 billion parameters when compared to previous-generation NVIDIA Hopper GPUs, even without specific Blackwell optimizations. The GB200 NVL72's architecture, featuring a large unified memory, FP4 precision, and a tightly integrated 72-GPU NVIDIA NVLink domain, allows every GPU to operate in concert. This interconnectedness results in dramatically higher throughput and reduced latency for token generation, leading to more performant and cost-effective inference. Autumn Moulder, vice president of engineering at Cohere, expressed enthusiasm about the seamless integration of their workloads onto the Grace Blackwell architecture, noting the "incredible performance efficiency across our stack" and anticipating further gains with ongoing optimizations.

Empowering Enterprise AI Models with IBM

IBM is leveraging the power of CoreWeave's extensive NVIDIA GB200 NVL72 deployment, which includes thousands of Blackwell GPUs, to train its next-generation Granite family of models. These models are designed to be open-source and enterprise-ready, focusing on high performance, safety, and cost-efficiency. The Granite models serve as the foundation for advanced solutions like IBM watsonx Orchestrate, a platform that empowers businesses to build and deploy AI agents capable of automating and accelerating enterprise workflows. Sriram Raghavan, vice president of AI at IBM Research, highlighted the excitement surrounding the acceleration that the NVIDIA GB200 NVL72 brings to their model training. He stated, "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." The integration also benefits from the IBM Storage Scale System, providing high-performance storage solutions tailored for large-scale AI workloads, further enhancing the capabilities available within CoreWeave's dedicated environments and AI cloud platform.

Advancing Open-Source Innovation with Mistral AI

Mistral AI, a prominent player in the open-source AI community based in Paris, is also harnessing CoreWeave's new infrastructure. The company has received its first thousand Blackwell GPUs, which are being used to develop the next generation of advanced language models. Mistral AI is leveraging the GB200 NVL72 systems, equipped with NVIDIA Quantum InfiniBand networking, to accelerate the development and deployment of models such as Mistral Large, known for its robust reasoning capabilities. Thimothee Lacroix, co-founder and CTO of Mistral AI, reported a significant performance improvement, noting a "2x improvement in performance for dense model training right out of the box and without any further optimizations." He further commented on the transformative potential of the NVIDIA GB200 NVL72, stating, "What’s exciting about NVIDIA GB200 NVL72 is the new possibilities it opens up for model development and inference." Mistral AI's adoption of this cutting-edge technology underscores the growing demand for scalable and high-performance computing resources in the open-source AI sector.

Scalability and Future Potential of Blackwell Instances

Beyond dedicated customer solutions, CoreWeave offers instances that feature rack-scale NVIDIA NVLink connectivity, encompassing 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs. These systems are interconnected via NVIDIA Quantum-2 InfiniBand networking and can scale up to an impressive 110,000 GPUs. This vast scalability, powered by the NVIDIA GB200 NVL72 rack-scale accelerated computing platform, provides the immense compute power required to build and deploy the next generation of AI reasoning models and intelligent agents. CoreWeave's infrastructure is designed to empower AI developers to innovate without limitations, offering the performance and scale necessary to tackle the most ambitious AI projects. The continuous expansion of Blackwell instances signifies a commitment to staying ahead of the curve in the rapidly evolving landscape of artificial intelligence, ensuring that customers have access to the most advanced tools for their AI endeavors.

The Transformative Impact on AI Development

The deployment of thousands of NVIDIA Grace Blackwell GPUs at CoreWeave marks a pivotal moment for the AI industry. By providing unparalleled performance, scalability, and efficiency, these systems are accelerating innovation across various domains, from personalized AI agents to complex enterprise models and open-source advancements. The collaboration between NVIDIA and CoreWeave is instrumental in democratizing access to cutting-edge AI hardware, enabling a wider range of researchers and developers to explore and implement advanced AI solutions. As more Blackwell-powered systems come online, the potential for transformative breakthroughs in artificial intelligence continues to grow, promising to reshape industries and redefine the capabilities of intelligent systems.