TLDR: CoreWeave, a prominent AI cloud-computing company, has launched its new AI Object Storage service, specifically engineered for artificial intelligence workloads. This fully managed service, powered by Local Object Transport Accelerator (LOTA) technology, promises instant global data accessibility without egress or transaction fees, scalable performance, and significantly reduced costs for AI training and inference.
LIVINGSTON, N.J. – October 24, 2025 – CoreWeave Inc. (Nasdaq: CRWV), a leading American AI cloud-computing company, has officially unveiled its groundbreaking AI Object Storage service. This new offering is meticulously designed from the ground up to cater specifically to the demanding requirements of artificial intelligence (AI) workloads, aiming to redefine how AI models access and scale data.
The CoreWeave AI Object Storage is a fully managed, S3-compatible service that distinguishes itself from traditional object storage systems by integrating tightly with GPU compute nodes. This integration is crucial for minimizing latency and maximizing throughput, which are critical factors for high-performance AI training and inference.
At the heart of this innovative storage solution is CoreWeave’s proprietary Local Object Transport Accelerator (LOTA) technology. LOTA deploys a proxy on each GPU node, enabling local caching of frequently used data and pre-staging objects on NVMe drives. This significantly reduces network traffic and ensures that a single dataset is instantly accessible anywhere in the world, eliminating egress charges and request/transaction fees.
Unlike conventional object storage, which is often confined to a single portion of an IT infrastructure, CoreWeave’s AI Object Storage performance scales dynamically as AI workloads grow. It maintains superior throughput across distributed GPU nodes, whether they are in any region, on any cloud, or on-premises. The system is fortified with private interconnects, direct cloud peering, and 400 GBps-capable ports, safeguarding the data integrity of trillions of objects for global workloads.
This multi-cloud networking backbone provides developers with consistent, high-throughput GPU performance, effectively eliminating challenges such as data sprawl and resource-intensive data replication. Morgan Fainberg, Principal Engineer at Replicate, highlighted the importance of this capability, stating, ‘With CoreWeave’s cross-cloud capabilities in CoreWeave AI Object Storage, we can rely on a single dataset to support models no matter where they’re deployed. This eliminates replication overhead, removes egress costs, and ensures our users always have high-performance access to the data they need to innovate.’
In terms of performance, benchmarks indicate that the system supports 2 GB/s per GPU throughput and is engineered to scale across large clusters of GPUs. Further technological advancements are expected to push throughput potential up to 7 GB/s per GPU in certain configurations. CoreWeave also emphasizes that its solution can cut storage costs for AI workloads by more than 75% compared to traditional storage systems.
The launch of AI Object Storage marks a strategic expansion for CoreWeave, extending its footprint beyond compute and GPU clusters into the data layer of AI systems. The company, established in 2017 and publicly listed on Nasdaq (CRWV) in March 2025, aims to position itself as a significant rival to established cloud providers like Amazon Web Services and Google Cloud. Following the announcement, CoreWeave’s shares (CRWV) saw a rise of approximately 2%, reflecting investor confidence in this infrastructure push.
Also Read:
- Amazon Web Services Intensifies AI Focus Amidst Rising Cloud Competition
- Amazon SageMaker HyperPod Enhances Distributed AI Model Training for Scaled Development
However, CoreWeave faces ongoing challenges, including a high level of debt and a significant increase in capital expenditure guidance for 2025, which could impact margins. The company also reported a net loss of $290.5 million in Q2 2025, raising questions about cost control. Dependence on a few major clients also remains a concern.
