Editorial illustration for NVIDIA's SCADA Server SDK Lets Partners Build Storage for GPU Clients
NVIDIA Opens GPU Storage Stack to Partners
NVIDIA's SCADA Server SDK Lets Partners Build Storage for GPU Clients
NVIDIA is opening up its storage stack for GPU clusters, giving partners a standard way to build fast, RDMA-based access to files and objects instead of stitching together vendor-specific code. The company announced general availability of cuObject client and server libraries on November 18, along with an expanded xio-sig specification that now covers cuObject in addition to cuFile. Google Cloud and Microsoft are named partners in the effort.
The problem NVIDIA is targeting is a familiar one for AI infrastructure teams: training and inference jobs need to pull data from object and file storage at speeds that match GPU throughput, using RDMA over ConnectX NICs or BlueField DPUs to skip the CPU memory copy entirely. Without a shared wire protocol for object storage, every storage provider has required its own integration path, slowing down engineers who just want their accelerators fed.
cuObject's APIs and RDMA wire protocol give developers a way to build storage servers and applications that talk the same language, while xio-sig lets that client work against any server implementation that follows the spec. A new SDK, aimed at storage partners, extends that interoperability further.
A new SCADA Server SDK enables storage providers to build SCADA servers that respond to GPU-initiated requests from SCADA clients. IBM has shown interoperability with a prototype that integrates SCADA and IBM Storage Scale. Together with the xio-sig expansion, these efforts give AI developers, storage providers, and cloud services more ways to build and use accelerated storage through shared APIs and protocols.
Why this matters
For anyone building AI infrastructure, storage has been the quiet bottleneck behind every GPU cluster's paper specs. NVIDIA opening up the SCADA Server SDK to storage partners is an admission that it can't own this layer alone, and that's the right call. If Pure Storage, DDN, VAST, or whoever else can build SCADA-compatible servers that talk RDMA natively to GPU clients, teams get to pick storage vendors on merit rather than settling for whatever ships bundled with the compute. That's good for founders watching infrastructure costs and good for researchers tired of storage I/O eating into training throughput.
The catch is standardization risk. An SDK controlled by NVIDIA means partners build to NVIDIA's roadmap, on NVIDIA's timeline, inside NVIDIA's definition of "fast enough." Watch whether cuObject and SCADA stay genuinely open to multiple storage backends or slowly become another lever that makes leaving the NVIDIA stack harder. The real test isn't the SDK announcement, it's which storage vendors actually ship compatible servers in the next year, and whether performance claims hold up outside NVIDIA's own benchmarks.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic NVIDIA NeMo powers telco reasoning model for autonomous network workflows
Industry Applications
From Archives Anthropic Warns Zhipu's GLM-5.3 Model Quickly Used to Build Exploits
From Archives Claude Sonnet 5.5 Review: Faster Agentic Coding & Visual QA
From Archives NVIDIA's AI Development Constraint: Isolation as a Scaling Unit
Research & Benchmarks
From Archives NVIDIA VSS Blueprint 3.3 Cuts Cost of Deploying Visual AI Agents
Common Questions Answered
What is the NVIDIA SCADA Server SDK and how does it benefit storage partners?
The NVIDIA SCADA Server SDK enables storage providers to build SCADA servers that respond to GPU-initiated requests from SCADA clients using standardized APIs and protocols. This eliminates the need for vendors to stitch together vendor-specific code, allowing storage partners like Pure Storage, DDN, and VAST to build RDMA-native solutions that compete on merit rather than being bundled with compute hardware.
What are cuObject client and server libraries and how do they relate to the xio-sig specification?
cuObject client and server libraries, which achieved general availability on November 18, provide a standard way for GPU clusters to access files and objects through fast, RDMA-based connections. The expanded xio-sig specification now covers cuObject in addition to cuFile, creating a unified standard that enables AI developers and storage providers to build accelerated storage solutions through shared APIs and protocols.
Which companies are named partners in NVIDIA's storage stack initiative?
Google Cloud and Microsoft are named partners in NVIDIA's effort to open up its storage stack for GPU clusters. Additionally, IBM has demonstrated interoperability with a prototype that integrates SCADA with IBM Storage Scale, showing how the SCADA Server SDK enables different storage vendors to achieve compatibility with GPU-initiated requests.
Why has storage been a bottleneck for GPU clusters according to this announcement?
Storage has traditionally been the quiet bottleneck behind GPU cluster performance because vendors had to implement custom, vendor-specific code rather than using standardized protocols. By opening the SCADA Server SDK, NVIDIA acknowledges it cannot own the storage layer alone and enables teams to select storage solutions based on merit rather than settling for bundled options that may not meet their performance requirements.
Further Reading
- Expanding AI Storage Access with NVIDIA cuObject and the NVIDIA SCADA Server SDK - NVIDIA Developer Blog
- Accelerating IO in the Modern Data Center: Magnum IO Storage Partnerships - NVIDIA Developer Blog
- GPUDirect RDMA and GPUDirect Storage - NVIDIA Docs
- NVIDIA GPUDirect - NVIDIA Developer
- NVIDIA-Certified Storage — NVIDIA RTX PRO AI Factory - NVIDIA Docs
