Targon Is Expanding From Confidential Compute Into a Full Cloud Stack With Bare Metal and Sandboxes

Most decentralized compute networks are still trying to prove they can supply enough GPUs. Targon is trying to solve a different problem.

Targon (SN4), Manifold Labs’ confidential compute infrastructure powered by Bittensor, is expanding beyond secure virtual machines with two new products called Bare Metal and Sandboxes.

With both products now live, the platform can serve teams that need confidential workloads, full physical hardware control, or fast temporary environments for development and testing.

Targon Is Expanding Beyond Confidential VMs

Targon (SN4) has mainly been known for confidential computing. Its network connects users to decentralized GPU and CPU resources while technologies like Intel TDX and NVIDIA Confidential Computing help keep workloads private.

Targon Virtual Machines, or TVMs, create protected environments where host operators cannot access customer data, model weights, or code. This makes them useful for sensitive workloads running on infrastructure the user does not fully trust.

But not every workload needs that level of isolation. Some teams want direct access to the hardware, while others need environments they can spin up quickly for testing, development, or agent tasks. Bare Metal and Sandboxes are built for those two different needs.

Bare Metal Gives Teams the Whole Machine

Bare Metal gives one customer exclusive access to an entire physical server, with no shared machine or virtualization layer in between. Teams can control the kernel, drivers, and GPU configuration directly, giving them more freedom to tune performance around the exact needs of their workload.

That makes it useful for model training, benchmarking, custom inference stacks, specialized drivers, and other workloads where direct hardware access is important. Its security model is different from Targon’s confidential virtual machines. Bare Metal does not use TVM cryptographic isolation and instead relies on KYC-verified operators.

The result is closer to a traditional dedicated server, but with the hardware still sourced through Targon’s decentralized compute marketplace. For teams that need maximum control, they get the whole machine and can configure it however they want.

Sandboxes Give Developers Environments in Seconds

Sandboxes take the opposite approach. They are isolated Linux environments built for speed and can be created in seconds for development, experiments, previews, or AI agent tasks.

Users can connect through a browser terminal, graphical Linux desktop, or SSH. A Sandbox can also be forked into multiple independent copies, allowing several workflows or agent tasks to run at the same time.

Configured environments can be saved as reusable templates, while idle instances can be paused so active compute billing stops until they are needed again. This setup is especially useful for AI agents that need somewhere to run commands, write code, inspect results, and recover when something fails.

Because each agent can work inside its own isolated environment, it can experiment without affecting production systems or other workloads. Developers can use the same setup for temporary projects and short-term development without maintaining a permanent machine.

Targon Now Covers Three Different Compute Needs

Targon now has three clear infrastructure paths.

1. Confidential virtual machines are built for sensitive workloads that need hardware-backed privacy.

2. Bare Metal is for teams that want full access to a physical machine and greater control over performance and configuration.

3. Sandboxes are for developers and agents that need fast, temporary environments they can create, copy, pause, and discard.

Each product serves a different type of workload, but they all sit inside the same platform. That is what makes the expansion more important than simply adding two new features.

Targon Is Starting to Look More Like a Full Cloud Platform

The bigger story is that Targon is moving beyond raw compute. AI teams need different kinds of infrastructure depending on the job, from dedicated hardware for training to confidential environments for sensitive data and lightweight Linux environments for agent workloads.

Targon is bringing those options into the same Bittensor-powered marketplace, turning decentralized compute into products that feel closer to the cloud tools developers already use. Manifold Labs has already been building in this direction through Targon’s confidential compute stack and its technical work with Intel. Bare Metal and Sandboxes extend that approach.

Targon is no longer offering only one way to run workloads on decentralized hardware. It is building a broader cloud layer where developers can choose the level of privacy, control, and flexibility their workload needs.