Confidential AI

Confidential AI has built the confidential computing stack for AI workloads that works at datacenter scale.

Confidential AI runs AI workloads inside trusted execution environments (TEEs): hardware-encrypted VMs on Intel TDX, AMD SEV-SNP and NVIDIA GPUs in confidential computing mode. Prompts, data and model weights stay invisible to the operator, us included. You don't have to trust us on that: the hardware signs an attestation report of exactly what code is running, and it is checked on your side before you send anything.

Built for AI labs, inference providers, routers and regulated companies.

Verify our inference cluster

What does Confidential AI sell?

We sell confidential CPU and GPU VMs, Confidential Kubernetes (C8s) built on them, and private inference built on C8s. Use our infra, or license our confidential computing stack for yours.

Confidential VMs

Rent CPU CVMs, full node or individual GPU CVMs. One confidential VM, yours for the term. GPUs in CC mode.

$68 per hour for an 8x B300 node on Intel TDX. B300 available now. B200 and H100 on request.

Pricing · Open the console

Confidential Inference

OpenAI-compatible endpoint. Open-weight models in TEEs. The model, the code and the hardware are verified on your side before the first token.

DeepSeek V4.1 Flash is live at $0.30 in and $1.20 out per 1M tokens. Kimi K3, GLM 5.3 and Qwen3.8 27B on request.

Pricing · Get an API key

Licensed

The full stack on your own GPUs: confidential VMs, Confidential Kubernetes (C8s), and private inference, agents and training on top. NVIDIA CC licensing included.

From $2,789 per GPU per year. Available now.

Pricing · Talk to us

Company

Based in San Francisco. Backed by Initialized Capital, Notable Capital and Soma Capital. Team · Careers

Member of

Confidential Computing Consortium, Linux Foundation, NVIDIA Inception and the NVIDIA Partner Network.

More

Blog · FAQ · Docs · Private AI · llms.txt