VAST Data launches DataEnclave for confidential AI execution

VAST Data has announced the launch of DataEnclave, a confidential AI capability integrated into its VAST DataEngine. Developed in collaboration with NVIDIA, the technology utilizes NVIDIA Confidential Computing to enable the execution of third-party AI models against sensitive enterprise data within secure, hardware-isolated environments.

DataEnclave is designed to resolve privacy conflicts in regulated sectors like healthcare and finance by allowing proprietary model weights and sensitive customer data to remain protected from infrastructure operators. The system uses cryptographic attestation to verify the execution environment before decrypting assets, ensuring that customer data keys and model keys remain under their respective owners' control. The solution supports NVIDIA's Hopper, Blackwell, and upcoming Rubin GPU platforms, encrypting CPU memory, GPU memory, and NVLink traffic.

Technical performance tests on NVIDIA Blackwell systems indicate that implementing these security features results in minimal overhead, maintaining approximately 96% to 98% of token throughput compared to unencrypted configurations. The ecosystem includes collaborations with hardware partners such as Cisco and Supermicro, as well as cloud providers like Sharon AI. While a commercial launch is expected in early 2027, the technology aims to transition AI models from external applications to managed, native resources within an integrated AI operating system.