Targon is a decentralised artificial intelligence (AI) infrastructure and confidential cloud computing platform developed by Manifold Labs. It operates as Subnet 4 (SN4) on the Bittensor network, providing a marketplace for computational resources and secure, scalable GPU and CPU rentals for training and deploying AI models. ^{[1] [2]}

Overview

Targon is designed to serve two primary functions: as a confidential cloud platform offering secure hardware rentals and as a decentralized compute market within the Bittensor ecosystem. The platform's core value proposition is to provide high-performance, cost-effective, and censorship-resistant AI compute services. It aims to enable the secure monetisation of proprietary AI models by ensuring their data and intellectual property remain protected during execution. ^{[3] [4]}

The platform's architecture is built around the Targon Virtual Machine (TVM), which leverages hardware-level security technologies to create isolated and verifiable environments for AI workloads. This allows developers to rent compute resources with cryptographic assurance that their models and data are not accessible, even by the infrastructure provider. Targon integrates OpenAI-compliant endpoints to simplify adoption for developers and supports various AI models, such as Llama-3 70b. The project's ecosystem includes distinct roles for miners, who provide computational resources, and validators, who ensure network integrity and service quality. ^{[5] [2]}

History

The development of Targon began with early commits to its public repository in 2024, with the project's license file being added on January 30, 2024. On March 31, 2025, Manifold Labs announced Targon v6, which detailed the architecture of the Targon Virtual Machine (TVM). ^{[5] [3]}

Throughout 2025, the project and its parent company, Manifold Labs, engaged in community-building and development activities. In early 2025, Manifold Labs hosted the first annual Bittensor Hackathon at ETHDenver, where teams competed to build applications using the Targon API. The company later announced "Manifold 2.0," stating its mission to build a decentralized frontier AI lab with Targon as a core component. On August 30, 2025, Manifold Labs announced it had raised a $10.5 million Series A funding round to accelerate the growth of the Targon compute market. ^[1]

The official X (formerly Twitter) account for the project, @TargonCompute, was created in June 2025. The platform's main product, "Secure Targon Compute," officially launched on October 28, 2025, enabling developers to rent H200 GPU and CPU nodes secured by the TVM. Following the launch, the team released a pre-configured Ubuntu image on October 31 and introduced support for attachable, persistent storage volumes on November 4, 2025. ^[6]

Technology

Targon's technology stack is centered on providing a secure and verifiable environment for AI computations through its proprietary Targon Virtual Machine and integration with the Bittensor network.

Core Architecture

Targon operates as Subnet 4 on the Bittensor network, a decentralised protocol that uses incentive mechanisms to coordinate a network of machine learning systems. Within this framework, Targon functions as a decentralized compute market where miners offer computational power and are rewarded based on their performance and reliability, which is assessed by validators. The platform utilizes InfiniBand for its networking infrastructure to facilitate low-latency performance for inference tasks. ^{[1] [2]}

Targon Virtual Machine (TVM)

The TVM is the core of Targon's security model, designed to deploy Confidential Virtual Machines (CVMs) on bare-metal hardware. It provides end-to-end security and attestation for both the host server and the attached GPUs. The TVM architecture consists of four primary layers:

• Host Verification: An initial process that evaluates a physical server's hardware for compatibility with confidential computing, checking for features like Intel TDX or AMD SEV support, a valid TPM 2.0 module, and Secure Boot configurations.
• Host Verification Attestation Service: A trusted service that validates the hardware report from the host. If the host meets security criteria, the service initiates the deployment of a CVM.
• CVM GPU Attestation: A service running inside the CVM that uses the NVIDIA nvTrust SDK to verify the integrity and authenticity of the NVIDIA GPU, generating a JWT-based attestation token as cryptographic proof.
• Manifold SDK: A software development kit that standardizes APIs and data models across all TVM components to ensure consistent communication and integration. ^[3]

Security Features

Targon's platform is built on a foundation of hardware-enforced security to protect data and models throughout the AI lifecycle.

• Confidential Computing: It utilizes hardware-based Trusted Execution Environments (TEEs) to protect data while it is in use. This is achieved through technologies like Intel Trust Domain Extensions (TDX) and AMD Secure Encrypted Virtualization (SEV).
• NVIDIA Integration: The platform integrates NVIDIA Confidential Computing to extend hardware-backed data protection to its GPU offerings, including support for Protected PCIe (PPCIE) to secure data in transit between the CPU and GPU.
• Remote Attestation: This process allows a user to cryptographically verify the hardware and software integrity of the compute environment before deploying any sensitive workloads.

The platform's roadmap includes plans for expanded functionality such as secure model training, multi-vendor support for technologies like AMD SEV-SNP, and the development of a comprehensive AI development platform with user-friendly tools. ^{[5] [2]}

Tokenomics

Targon's native token is identified by the ticker symbol SN4, signifying its role as Bittensor's Subnet 4.

• Token Name: Targon
• Ticker: SN4
• Network: Bittensor
• Max Supply: 21,000,000 SN4
• Total Supply: As of November 2025, the total supply was reported to be approximately 3,129,468 SN4.

The SN4 token is used within the Targon subnet's incentive mechanism to reward miners for providing compute resources and validators for ensuring network integrity. The token is traded on decentralized exchanges, with the primary marketplace being Subnet Tokens. The most active trading pair is SN4 against TAO, the native token of the Bittensor network. ^{[7] [8]}

Team and Funding

Targon is developed and maintained by Manifold Labs Inc., an AI infrastructure company headquartered in Austin, Texas. The company specializes in decentralized machine learning systems and Bittensor integration. ^[1]

Key figures at Manifold Labs include:

• Robert Myers: Founder and CEO. Myers was previously a Senior Software Engineer at the Opentensor Foundation, where he contributed to the development of Bittensor.
• James Woodman: Co-founder.
• Viraj Sahu: An individual associated with the project, suggested to be a founder or key team member. ^{[1] [6]}

Funding

On August 30, 2025, Manifold Labs announced it had closed a $10.5 million Series A funding round. The round was led by OSS Capital, with participation from Digital Currency Group (DCG), Tobias Lütke, Ram Shriram, and others. The company stated that the funds would be used to accelerate the growth of the Targon compute market and its cloud inference business. ^[1]

Use Cases and Products

Targon's infrastructure is designed to support a range of AI-related services, primarily focused on AI inference, model leasing, and general-purpose GPU compute. The platform offers a "Playground" interface for users to test and compare different large language models in real-time. ^[4]

Manifold Labs has developed several applications on top of the Targon network to demonstrate its capabilities:

• Dippy: An application for AI-driven character chats.
• TAOBOT: A tool that serves as an access point to various decentralized AI services.
• Sybil: An AI-enhanced search tool.

These products showcase the platform's ability to support user-facing applications in sectors such as entertainment and information retrieval. ^[4]

Criticisms

A research report published by Token Metrics in May 2025 highlighted several areas where the project lacked transparency at the time. The report noted that the project did not prominently disclose information about its core team, which could limit verifiability for users and investors. It also pointed out that public GitHub repositories had limited visibility, making a comprehensive independent code audit difficult. Finally, the report mentioned the absence of a formal, detailed public roadmap, which made it challenging to assess the project's long-term strategic direction. ^[4]