Confidential AI

Confidential AI runs AI workloads inside trusted execution environments (TEEs), using hardware-encrypted virtual machines on Intel TDX, AMD SEV-SNP and NVIDIA GPUs in confidential computing mode. Prompts, data and model weights are not visible to the operator, and clients verify remote attestation before sending any data. The company lets users verify its production cluster in a browser. Offerings: - Confidential Inference: an OpenAI-compatible endpoint serving open-weight models inside TEEs, priced per million tokens. The client verifies the model, code and hardware before the first token. DeepSeek V4 Flash is live, with other models available on request. - Confidential VMs: rentable confidential GPU virtual machines (B300, B200, with H100 on request), either full nodes or individual GPUs, with root SSH access and GPUs in confidential computing mode. - Licensed stack: software for customers' own GPUs, including Confidential Metal, Confidential Kubernetes, inference services, Confidential OS Builder and SDKs, with NVIDIA confidential computing licensing included. It is priced per GPU per year. The target customers are AI labs, inference providers, routers and regulated companies. The company is based in San Francisco and is backed by Initialized Capital, Notable Capital and Soma Capital. It is a member of the Confidential Computing Consortium, the Linux Foundation, Nvidia Inception and the Nvidia Partner Network.