NVIDIA Confidential Computing
NVIDIA Confidential Computing secures data in use, protecting AI models and workloads as they execute, by leveraging hardware-based trusted execution environments built into NVIDIA Hopper and Blackwell architectures and supported platforms. It enables enterprises to deploy AI training and inference, whether on-premises, in the cloud, or at the edge, with no changes to model code, while ensuring the confidentiality and integrity of both data and models. Key features include zero-trust isolation of workloads from the host OS or hypervisor, device attestation to verify that only legitimate NVIDIA hardware is running the code, and full compatibility with shared or remote infrastructure for ISVs, enterprises, and multi-tenant environments. By safeguarding proprietary AI models, inputs, weights, and inference activities, NVIDIA Confidential Computing enables high-performance AI without compromising security or performance.
Learn more about NVIDIA Confidential Computing
Tinfoil
Tinfoil is a verifiably private AI platform built to deliver zero-trust, zero-data-retention inference by running open-source or custom models inside secure hardware enclaves in the cloud, giving you the data-privacy assurances of on-premises systems with the scalability and convenience of the cloud. All user inputs and inference operations are processed in confidential-computing environments so that no one, not even Tinfoil or the cloud provider, can access or retain your data. It supports private chat, private data analysis, user-trained fine-tuning, and an OpenAI-compatible inference API, covers workloads such as AI agents, private content moderation, and proprietary code models, and provides features like public verification of enclave attestation, “provable zero data access,” and full compatibility with major open source models.
Learn more about Tinfoil
Inkling
Inkling is an open-weights multimodal AI model from Thinking Machines designed as a customizable foundation model for developers, researchers, and enterprises. The model is a Mixture-of-Experts transformer with 975 billion total parameters, 41 billion active parameters, and support for context windows up to 1 million tokens. Inkling was trained from scratch on text, images, audio, and video, giving it native capabilities across reasoning, coding, agentic tool use, vision, audio, factuality, and instruction following. It is built with controllable thinking effort so users can balance performance, latency, and token efficiency for different workloads. The model is available for fine-tuning on Tinker, with playground access, API availability through ecosystem partners, and full weights published on Hugging Face. Built for customization, Inkling gives teams an open-weights base model for building domain-specific AI systems, multimodal agents, coding workflows, research tools, and more.
Learn more about Inkling
Confidential AI
Confidential AI provides private inference, private agents, confidential VMs and confidential computing infrastructure for AI labs, inference providers, routers, and organizations handling sensitive workloads. Its hosted inference service runs open-weight models in hardware-backed trusted execution environments (TEEs), including Intel TDX and AMD SEV-SNP confidential VMs with NVIDIA GPUs in confidential computing mode. Prompts, responses, and workload data are protected from infrastructure operators, and clients can verify attestation before sending sensitive input. Customers can use an OpenAI-compatible API to deploy inference endpoints and agent workloads, or license the confidential-computing stack for on-premises infrastructure, bare metal, and major clouds. Use cases include serving models while limiting operator access to prompts and model assets, and running sensitive AI workloads on dedicated confidential compute.
Learn more about Confidential AI