Vast Data launches confidential computing service for sensitive workloads

Vast Data Inc., a company specializing in artificial intelligence technology, has unveiled DataEnclave, a confidential computing environment. This new feature, integrated within the Vast DataEngine and the Vast AI Operating System, employs Nvidia's confidential computing technology to establish hardware-segregated spaces. Within these secure environments, proprietary models and enterprise data can be decrypted and processed safely.

The service is aimed at entities like financial institutions, healthcare providers, and government agencies that cannot transfer regulated or sensitive information to AI services hosted externally. It also addresses model developers' concern over distributing proprietary model weights to infrastructure they do not control. Jeff Denworth, co-founder of Vast Data, highlighted that model weights are increasingly becoming the most valuable intellectual property worldwide.

DataEnclave utilizes trusted execution environments across CPUs and GPUs. It encrypts memory, GPU memory, and NVLink traffic while isolating active models and data from administrators, operators, and other tenants. A cryptographic attestation process validates the hardware, software, and policies before decrypting models or data. Enterprises and developers retain control over their encryption keys via integrations with their own key management systems.

Denworth noted that until now, there was no adequate solution for ensuring that the computing environment's memory space is equally protected. Once computing reaches its destination, the security of the environment becomes crucial for decrypting sensitive data. DataEnclave can be deployed in customer data centers, AI clouds, and air-gapped environments. Its attestation services are based on the Cloud Native Computing Foundation Trustee project or can be provided through Fortanix Inc.'s Confidential AI offering.

The system logs attestation events, key releases, and enclave lifecycle activity in the Vast DataBase data warehouse, providing an audit trail of workloads without disclosing the protected data or model weights. Initial partners include Cohere, CrowdStrike, Deepgram, The San Francisco AI Factory, Fundamental Research Labs, and TwelveLabs. Hardware partners include Cisco and Super Micro Computer.

Denworth emphasized that this setup allows developers' base weights and customers' fine-tuned weights to operate together under separate ownership and key control. This could help enterprises create customized models without sharing proprietary training or operational data with base-model providers. Initially, Vast targets organizations with stringent security and data residency requirements, such as large banks, intelligence agencies, and pharmaceutical companies.

The company believes that DataEnclave could alter the economics of enterprise AI. Enterprises could run proprietary models on their infrastructure or select a regional AI cloud, reducing the need for developers to finance underlying computing capacity. Vast plans to integrate DataEnclave into its existing AI Operating System, making it available in preview and scheduled for release in Q1 2027 through Vast and hardware partners like Cisco and Supermicro.