AI hyperscaler CoreWeave has officially launched access to NVIDIA’s powerful GB200 NVL72 rack-scale systems, naming IBM, Cohere, and Mistral AI as its first customers to leverage the cutting-edge infrastructure. The deployment marks a significant milestone in the evolution of AI cloud services, combining NVIDIA’s Grace Blackwell Superchips with CoreWeave’s full suite of performance-optimized cloud technologies.
The aim is to accelerate the development and deployment of next-generation AI models, especially those focused on reasoning and agentic capabilities.
CoreWeave’s platform is purpose-built for speed, and this latest launch reaffirms the company’s reputation as a pioneer in operationalizing the most advanced computing systems. “CoreWeave is built to move faster – and time and time again, we’ve proven it,” said Michael Intrator, co-founder and CEO of CoreWeave. “Today’s announcement demonstrates our engineering prowess and unwavering focus on supporting the next wave of artificial intelligence. We’re thrilled to see some of the most innovative companies in AI use our infrastructure to push boundaries and build what was previously impossible.”
The GB200 NVL72 systems incorporate NVIDIA’s Grace Blackwell architecture, which is engineered specifically for AI reasoning and agentic workloads. These systems are integrated into CoreWeave’s infrastructure, which also includes Kubernetes-native services such as CoreWeave Kubernetes Service, Mission Control, and Slurm on Kubernetes (SUNK). The result is a flexible and scalable environment tailored to handle the increasing complexity of enterprise AI applications.
“Enterprises worldwide are racing to turn reasoning models into agentic AI applications that will transform how people work and live,” said Ian Buck, Vice President of HPC and Hyperscale at NVIDIA. “CoreWeave’s rapid deployment of GB200 systems is laying the foundation for AI factories to become a reality.”
Scalability to 110,000 Blackwell GPUs
The company’s efforts are backed by performance data. In the latest MLPerf v5.0 benchmarking tests, CoreWeave set a new record in AI inference using the NVIDIA GB200 Grace Blackwell Superchips. These tests provide industry-standard metrics for evaluating the practical performance of machine learning workloads in real-world conditions.
The GB200-powered systems are connected via NVIDIA’s Quantum-2 InfiniBand, enabling scalability to as many as 110,000 Blackwell GPUs. This architecture supports the demands of modern AI applications, offering both the performance and reliability required by developers and enterprise AI labs.
CoreWeave has also formed notable strategic relationships, including a recently announced multi-year partnership with OpenAI. This adds to an expanding list of high-profile clients such as IBM, Mistral AI, and Cohere, all of whom are now able to take advantage of the GB200 NVL72 infrastructure to build and scale advanced AI models.
Related News
Here are three related articles on HostingJournalist.com:
-
CoreWeave Claims AI Inference Record with NVIDIA GB200This article highlights CoreWeave achieving record AI inference speeds using NVIDIA GB200 chips, surpassing previous benchmarks with 800 TPS on Llama 3.1 models, showcasing the performance capabilities of the GB200 NVL72 platform. Read article.
-
CoreWeave Partners with Bulk for Major NVIDIA AI Deployment in EuropeIt covers CoreWeave’s collaboration with Bulk to deploy a large-scale NVIDIA GB200 NVL72 cluster in Europe, supported by NVIDIA Quantum-2 networking, expanding the reach of this advanced AI infrastructure. Read article.
-
CoreWeave and Dell Technologies Expand Partnership to Scale AI SolutionsThis article details CoreWeave’s expanded partnership with Dell Technologies to deliver customized rack systems powered by NVIDIA GB200 NVL72, emphasizing energy efficiency and power management for AI workloads at scale. Read article.
