CoreWeave Deploys NVIDIA Vera Rubin

CoreWeave has achieved a significant milestone by becoming the first AI cloud provider to successfully bring up and validate the NVIDIA Vera Rubin NVL72 system. This deployment positions the company at the forefront of offering access to NVIDIA’s most advanced AI hardware, crucial for powering the next generation of agentic AI applications.

The Vera Rubin NVL72 represents a leap forward in AI compute, featuring 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs per rack. Connected by a 260 TB/s NVIDIA NVLink 6th-generation fabric, this system promises up to 10 times better inference per watt, requires fewer GPUs, and offers a significantly lower cost per million tokens compared to the previous NVIDIA Blackwell platform. According to the announcement, this translates directly into better results for CoreWeave’s customers.

"Our research depends on infrastructure that's both powerful and reliable, and CoreWeave has delivered on this as we've scaled across NVIDIA Hopper and Blackwell," said Craig Falls, head of Quantitative Research at Jane Street. "Their ability to deliver highly performant clusters with full cluster observability and a support team that engages deeply on hard problems gives us the confidence to partner with them on Vera Rubin."

Purpose-Built Infrastructure for Rack-Scale AI

To fully exploit the capabilities of Vera Rubin at scale, CoreWeave has introduced several proprietary infrastructure management innovations. These include Software-Defined Liquid Cooling through its Valvey system, which transforms cooling into a programmable, rack-level control surface. Valvey monitors and manages cooling systems in real time, enabling automated responses to issues without disrupting other racks.

Additionally, CoreWeave developed Racky, a unified rack control appliance that standardizes the management of power, cooling, and environmental sensors. This allows each Vera Rubin rack to be treated as a cloud resource. The platform also supports advanced networking with NVIDIA Quantum-X800 InfiniBand and Spectrum-X Ethernet, delivering 1.6 Tb/s of backend bandwidth per GPU and scaling to hundreds of thousands of GPUs. The integration of NVIDIA BlueField-4 DPUs further enhances security and tenant isolation, accelerating data access and lowering latency for multi-tenant operations.

Chen Goldberg, executive vice president of Product & Engineering at CoreWeave, highlighted the shift in infrastructure demands. "The agentic era demands a fundamentally different approach to infrastructure, one that keeps pace with workloads that reason continuously, scale unpredictably, and operate in production around the clock," she stated. "What separates infrastructure that performs in a lab from infrastructure that performs in production is the depth of engineering underneath it."

Industry Context and Competitive Positioning

CoreWeave’s strategic move to deploy the latest NVIDIA hardware underscores the intense competition in the AI cloud infrastructure market. Companies are vying to offer the most performant and cost-effective solutions for increasingly demanding AI workloads. This deployment of Vera Rubin is a direct response to the growing need for specialized compute for large-scale model training and, critically, for inference at the scale required by agentic AI systems, which demand persistent reasoning and massive context windows.

While CoreWeave is making waves, the AI infrastructure space is crowded. Competitors like Nebius (StartupHub score 85/100) and Applied Digital (StartupHub score 70/100) are also aggressively expanding their GPU capacity. CoreWeave, with a StartupHub score of 67/100 and verified financials showing it raised $900M via a junk-bond sale in 2026, is positioning itself as the essential cloud for AI pioneers. Its focus on purpose-built infrastructure, as demonstrated with Vera Rubin, aims to differentiate it from more general-purpose cloud providers.

The collaboration with hardware partners is also key. Dell Technologies provided the server backbone with its PowerEdge XE9812 servers, and Micron contributed high-performance SSDs. This integration highlights the complex supply chain and engineering effort required to bring such advanced systems to market. Ian Buck, vice president of Hyperscale and HPC at NVIDIA, noted, "CoreWeave has consistently been at the frontier of deploying each new generation of NVIDIA architecture at scale."

Why This Matters for AI Developers and Enterprises

For AI developers and enterprises, this means access to potentially transformative compute power. The efficiency gains promised by Vera Rubin could significantly accelerate the development and deployment of complex AI models and applications. The focus on inference performance is particularly relevant as AI agents become more sophisticated and are expected to operate continuously in production environments. This deployment by CoreWeave could lead to faster iteration cycles, reduced operational costs, and the ability to tackle previously intractable AI problems.

The validation of rack-scale systems like Vera Rubin is crucial. It moves beyond theoretical performance to demonstrate real-world capability. Companies that can offer reliable, high-performance access to this hardware, coupled with sophisticated management tools, will be well-positioned to capture market share. This news also signals the ongoing importance of deep partnerships between hardware manufacturers like NVIDIA and specialized cloud providers like CoreWeave to push the boundaries of AI compute.