CoreWeave Brings Up Multi-Rack NVIDIA Vera Rubin NVL72 Cluster

CoreWeave adds multi-rack NVIDIA Vera Rubin NVL72 clusters and new storage features to speed agentic AI and lower long-term data costs.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

New capabilities across compute, storage, and networking keep customers productive and accelerate the AI loop

LIVINGSTON, N.J.--(BUSINESS WIRE)-- CoreWeave, Inc. (Nasdaq: CRWV), The Essential Cloud for AI™, today announced the bring up of multi-rack NVIDIA Vera Rubin NVL72 on CoreWeave Cloud, putting hundreds of NVIDIA Rubin GPUs into a single scale-out cluster for agentic AI. The company also announced two new capabilities in CoreWeave AI Object Storage, cross-region write acceleration and a new Archive tier, that keep the data those workloads depend on close to the GPUs.

Multi-rack Vera Rubin NVL72 clusters allow for training and inference jobs to run across hundreds of Rubin GPUs. That matters for multi-step agentic workloads, which are sensitive to data-access latency because delays can compound across repeated model calls and tool use. Cross-region write acceleration removes the wait even when working across multiple regions. A job writes locally while CoreWeave replicates the data to another region in the background, so an agent's intermediate state, retrieved context, and outputs move as fast as its reasoning. The GPUs don't wait, and neither does the loop.

“CoreWeave was the first AI cloud provider to validate and bring up a Vera Rubin NVL72, demonstrating that this advanced rack-scale architecture could operate as a reliable, high-performance cloud service,” said Chen Goldberg, executive vice president of product & engineering at CoreWeave. “With multi-rack Vera Rubin, we are connecting hundreds of Rubin GPUs as a single scale-out cluster. For customers building agentic AI, that means greater scale, faster iteration, and higher productivity as models and agents continuously learn and improve.”

Scaling agentic AI with multi-rack NVIDIA Vera Rubin NVL72

A single NVIDIA Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs, NVIDIA NVLink 6, NVIDIA ConnectX-9 SuperNICs, and NVIDIA BlueField-4 DPUs. With multi-rack Vera Rubin NVL72, CoreWeave unifies racks of hundreds of accelerators using NVIDIA Spectrum-X Ethernet networking into a single scale-out cluster. This delivers the capacity to train larger models, serve more demanding inference workloads, and run reinforcement learning at scale, and achieves required engineering at every layer across compute, networking, storage, cooling, power, firmware, and software, to perform as one coordinated system.

CoreWeave brings multiple racks up as a single system through:

• Automating rack life cycle control. Racks arrive as hardware that needs to be connected and validated. CoreWeave Mission Control^® automates rack setup through the Rack LifeCycle Controller, which coordinates hardware detection, firmware updates, validation, power, and cooling, with Racky providing rack control and Valvey executing cooling actions.
• Validating performance from components to systems. CoreWeave combines NVIDIA field diagnostics with full-rack workload testing, comparing every result, building upon years of real-world experience, before a rack goes into production. Only racks that clear this bar as a system move into production, so every GPU performs at its best.
• Scaling the network with the GPUs. Every Rubin GPU is equipped with two NVIDIA Connect X-9 SuperNICs per Rubin GPU, providing 1.6 Tb/s of connectivity scale out connectivity per GPU across multiplane, multirail paths, supporting roughly 128,000 GPUs per rail in a non-blocking fabric. The modular topology allows racks to be added without redesigning the fabric at each expansion.

Keeping AI data close to compute across regions

AI performance depends on keeping data moving as quickly as compute. CoreWeave AI Object Storage LOTA (Local Object Transport Accelerator) brings data closer to AI workloads through managed caching on each CoreWeave Kubernetes Service node, delivering reads at local NVMe speeds and reducing latency by 8x compared to reading from a traditional storage cluster. LOTA provides up to 7 GB/s of throughput per GPU and scales linearly as clusters grow, helping eliminate network bottlenecks and accelerate AI training.

“Our datasets span multiple regions, and we can’t afford to have our training schedule dictated by cross-region retrieval delays,” said Cécile Robert-Michon, director of internal infrastructure at Cohere. “CoreWeave AI Object Storage gives us a unified dataset footprint across regions with reads cached locally, so nothing waits on the network. It’s the difference between planning around our data and simply training.”

New capabilities now allow customers to run workloads across two regions while avoiding cross-region write latency and reducing long-term storage costs.

• Checkpoints can now be written locally within a region, closest to the available compute. Data can now be written at local latency while being migrated to a second remote region in the background. This benefits companies with compute in multiple regions, but data in only one region, that are writing checkpoints during training jobs. The new cross-region write feature minimizes the pause, and enables the CKS cluster to continue training. Since the application sees a single bucket, there are no code changes, and permissions and retention rules work the same regardless of which region a write came from. This improves operations by eliminating the need to manually copy or move data from one region to another for critical AI workloads.
• The decision to keep critical data longer just became easier and less expensive. Teams routinely delete data they would rather keep: the checkpoint from the run that almost worked, the dataset needed to reproduce a result, the model version someone may ask about in six months. Archive is a new lower-cost storage tier built for exactly that data, with no fee to retrieve it, no fee to delete it early, and no fee when reading it from the Archive tier.

CoreWeave consistently delivers industry-leading performance, demonstrated by record-breaking MLPerf benchmark results in inference and training, its position as the only AI cloud to earn the top Platinum ranking in both SemiAnalysis ClusterMAX™ 1.0 and 2.0, and its #1 ranking for inference speed and price-performance for Moonshot AI’s Kimi K2.6 and Kimi K2.7 Code in independent inference benchmarking conducted by Artificial Analysis.

About CoreWeave

CoreWeave is The Essential Cloud for AI™. Built for pioneers by pioneers, CoreWeave delivers a platform of technology, tools, and teams that enables innovators to move at the pace of innovation, building and scaling AI with confidence. Trusted by leading AI labs, startups, and global enterprises, CoreWeave serves as a force multiplier by combining superior infrastructure performance with deep technical expertise to accelerate breakthroughs. Established in 2017, CoreWeave completed its public listing on Nasdaq (CRWV) in March 2025. Learn more at www.coreweave.com.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260916102944/en/

press@coreweave.com

Source: CoreWeave, Inc.

Key Terms

inference technical

Inference is the process of drawing a conclusion from available evidence or data, like a detective piecing together clues to form a likely story. For investors it matters because these judgments turn raw reports, test results, or market signals into expectations about future performance, risk, or regulatory outcomes—so how someone infers from the same facts can change investment decisions and valuation.

reinforcement learning technical

A type of artificial intelligence that learns by trial and error, receiving feedback from its actions to favor choices that lead to better outcomes. Think of it like a salesperson learning which pitches close deals by trying different approaches and keeping the ones that work. For investors, reinforcement learning matters because it can power smarter trading systems, optimize business operations, or improve products—potentially boosting efficiency and profits while also introducing model and execution risks.

nvme technical

NVMe is a fast data-transport standard that lets modern solid-state drives (SSDs) move information much more quickly and efficiently than older interfaces, acting like a wider, faster highway between storage and a computer’s processor. For investors, NVMe matters because it boosts device and server performance, can lower operating costs and power use in data centers, and influences which products and suppliers are competitive in markets where speed and efficiency drive revenue and margins.

kubernetes technical

Kubernetes is an open-source system that automates running and managing many pieces of software across groups of computers, like a conductor coordinating musicians so each piece plays at the right time and place. For investors, it matters because companies that use it can deploy updates faster, scale services up or down automatically, and cut infrastructure costs — factors that influence growth, reliability and operating margins.