Today’s AI workloads are data-intensive, requiring more scalable and affordable storage than ever. By 2028, enterprises are projected to generate nearly 400 zettabytes of data annually, with 90% of new data being unstructured, comprising audio, video, PDFs, images and more.
This massive scale, combined with the need for data portability between on-premises infrastructure and the cloud, is pushing the AI industry to evaluate new storage options.
Enter RDMA for S3-compatible storage — which uses remote direct memory access (RDMA) to accelerate the S3-application programming interface (API)-based storage protocol and is optimized for AI data and workloads.
Object storage has long been used as a lower-cost storage option for applications, such as archive, backups, data lakes and activity logs, that didn’t require the fastest performance. While some customers are already using object storage for AI training, they want more performance for the fast-paced world of AI.
This solution, which incorporates NVIDIA networking , delivers faster and more efficient object storage by using RDMA for object data transfers.
For customers, this means higher throughput per terabyte of storage, higher throughput per watt, lower cost per terabyte and significantly lower latencies compared with TCP, the traditional network transport protocol for object storage.
Lower Cost: End users can lower the cost of their AI storage, which can also speed up project approval and implementation.
Workload Portability: Customers can run their AI workloads unmodified in both on premises and in cloud service provider and neocloud environments, using a common storage API.
Accelerated Storage: Faster data access and performance for AI training and inference — including vector databases and key-value cache storage for inference in AI factories.
AI data platform solutions gain faster storage object storage access and more metadata for content indexing and retrieval.
Reduced CPU Utilization: RDMA for S3-compatible storage doesn’t use the host CPU for data transfer, meaning this critical resource is available to deliver AI value for customers.
NVIDIA has developed RDMA client and server libraries to accelerate object storage. Storage partners have integrated these server libraries into their storage solutions to enable RDMA data transfer for S3-API-based object storage, leading to faster data transfers and higher efficiency for AI workloads.
Client libraries for RDMA for S3-compatible storage run on AI GPU compute nodes. This allows AI workloads to access object storage data much faster than traditional TCP access — improving AI workload performance and GPU utilization.
While the initial libraries are optimized for NVIDIA GPUs and networking, the architecture itself is open, because other vendors and customers can contribute to the client libraries and incorporate them into their software. They can also write their own software to support and use the RDMA for S3-compatible storage APIs.
NVIDIA is working with partners to standardize RDMA for S3-compatible storage.
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://blogs.nvidia.com/blog/s3-compatible-ai-storage/#content
- https://www.nvidia.com/en-us/
- https://blogs.nvidia.com/?s=
- AMD's FSR Redstone officially debuts in Call of Duty: Black Ops 7 — AMD's answer to Nvidia's Ray Reconstruction
- How to Unlock Accelerated AI Storage Performance With RDMA for S3-Compatible Storage
- Microsoft to appeal ruling in favor of reselling perpetual Windows licenses — UK Competition Court says fineprint holds no ground as judges throw out company's
- Centauri Carbon 2 spotted in the wild ahead of Q1 2026 release – multi-material 3D printer's details listed in first official spec page
- Memory makers have no plans to increase RAM production despite crushing memory shortages — 'modest' 2026 increase predicted as DRAM makers hedge their AI bets
Informational only. No financial advice. Do your own research.