CoreWeave and Nvidia Validate Vera Rubin NVL72 for Agentic AI
CoreWeave and Nvidia have validated the Vera Rubin NVL72 rack-scale system, designed to support production-grade agentic AI workloads with high-bandwidth NVLink 6 connectivity. The architecture, supported by Dell PowerEdge servers, leverages specialized rack-control appliances and liquid cooling to manage large-scale inference tasks.
Key Takeaways
- Vera Rubin NVL72 integrates 72 Rubin GPUs and 36 Vera CPUs into a single rack-scale compute unit.
- Internal rack bandwidth reaches 260 TB/s via NVLink 6, exceeding the data throughput of the global internet.
- New 'Valvey' and 'Racky' hardware appliances provide software-defined liquid cooling and unified telemetry management.
- Networking stack supports Nvidia BlueField-4 DPUs alongside Spectrum-X Ethernet and Quantum-X800 InfiniBand rails.
Why It Matters
The validation marks a shift from isolated GPU servers to rack-as-a-computer architectures necessary for agentic AI, which requires non-stop multi-step reasoning. For the streaming and media ecosystem, this infrastructure provides the high-bandwidth, low-latency backbone required for real-time generative video and autonomous metadata tagging at scale. By treating the entire rack as a single pool of memory and compute, CoreWeave reduces the performance bottlenecks that previously limited trillion-parameter model inference. Watch for CoreWeave to integrate more autonomous training features via its Weights & Biases acquisition to further automate infrastructure scaling.
Additional Context
The rollout of Nvidia’s Vera Rubin architecture follows the company’s aggressive shift to a one-year product release cycle, a strategy CEO Jensen Huang confirmed during the June 2024 Computex keynote. This accelerated roadmap aims to maintain Nvidia's lead in the AI infrastructure market, which Gartner projected would reach $163 billion by the end of 2026. Per Bloomberg, June 2026, CoreWeave has secured over $12 billion in recent financing to build out these specialized data centers, positioning itself as a primary alternative to hyperscalers like AWS for high-density GPU workloads. Simultaneously, Dell Technologies has reported record demand for its AI-optimized server line. According to Dell's Q1 2026 earnings report, its AI server backlog reached nearly $4 billion, driven largely by the XE series validated in this CoreWeave partnership. Analysts at Omdia noted in May 2026 that the shift toward liquid-cooled, rack-scale systems is no longer optional for enterprises running Blackwell and Rubin-class chips, as power densities per rack now frequently exceed 100kW. This thermal demand is forcing a significant retrofit of legacy data center facilities globally to accommodate the specialized cooling units like CoreWeave's Valvey system.
Read full article at siliconangle.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source