NVIDIA announced on September 30, 2026 that its cuObject client and server libraries are generally available. The libraries provide APIs and a remote direct memory access, or RDMA, wire protocol for accelerated access to object storage. NVIDIA is also expanding xio-sig to include cuObject alongside cuFile.
The same technical post announces a Scaled Accelerated Data Access (SCADA) Server SDK. NVIDIA says storage providers can use it to build servers that respond to GPU-initiated requests from SCADA clients and return results over RDMA.
Availability of libraries is not the same as completion of every interoperability effort. NVIDIA says headers, wire-protocol material and implementation code will be shared after the production-ready stack passes conformance tests. Its post says governance documents are still under review by pending board members.
NVIDIA reports that IBM demonstrated a prototype integrating SCADA with IBM Storage Scale. A prototype demonstration is not proof of a finished cross-vendor standard or measured gains for every storage workload.
For developers, the update is a new set of storage-access tools and ongoing interoperability work, rather than a blanket promise that existing AI storage bottlenecks have disappeared.
Source: https://developer.nvidia.com/blog/expanding-ai-storage-access-with-nvidia-cuobject-and-the-nvidia-scada-server-sdk/