Nvidia opens cuFile to four maintainers

Nvidia has open-sourced its cuFile APIs and underlying storage stack under an inaugural group of 4 maintainers: Google, Intel, Meta, and Nvidia. The company’s FMS announcement says cuFile lets GPUs read from and write to storage directly, while the new XIO organization describes a vendor-neutral stack spanning four repositories. For infrastructure teams already modeling flash as a second tier beside accelerator memory, the practical signal is interoperability: test whether the open conformance layer can reduce custom storage integration before treating GPU-direct I/O as portable infrastructure.