Table of Contents
Subnet 4 Targon, the compute platform operated by Manifold, has launched Bare Metal and Sandboxes, adding two new infrastructure products to its decentralized cloud marketplace.
The release, announced during Exploit Summit in Montreal, expands Targon beyond confidential container workloads into dedicated physical machines and short-lived Linux environments. Together, the products target two different ends of the compute market: teams that need direct control over full hardware and developers that need disposable environments for agents, previews, testing, and experimentation.
The Manifold team is enjoying Montreal at Exploit Summit, where we're announcing new features and roadmaps.
— Manifold (@manifoldlabs) September 29, 2026
As part of our new release rollout, we've launched Bare Metal and Sandboxes on Targon: a full physical machine when you need the whole box, and disposable Linux… pic.twitter.com/9SovoAhXRp
The launch also broadens the commercial surface area for one of Bittensor's earliest compute subnets. Targon has focused on confidential GPU compute, with its Targon Virtual Machine model routing work across verified providers. Bare Metal and Sandboxes give the network more ways to serve workloads that do not fit cleanly into a single containerized deployment model.
Bare Metal Gives Users the Whole Machine
Targon's Bare Metal product gives customers access to an entire physical machine rather than a virtualized slice of shared infrastructure. Users get no hypervisor, no container runtime, and no noisy neighbors, along with full control over the kernel, drivers, and firmware-level GPU settings.
That setup suits workloads where abstraction layers can become constraints. Teams working with custom drivers, low-level GPU tuning, specialized inference stacks, or sensitive performance requirements often need more control than a managed container or virtual machine can provide. Bare Metal is Targon's answer for those users.
Targon said Bare Metal operators are KYC-verified, which the team framed as a way to preserve workload security without requiring a Trusted Virtual Machine. The model shifts the product from cryptographic isolation around a confidential container toward a more conventional dedicated-server relationship, while still routing supply through Targon's Bittensor-based compute marketplace.
A dedicated physical server can be useful for AI teams when a workload needs predictable performance, full hardware access, or deeper control over the software stack. It is less flexible than a disposable environment, but it can be more appropriate for high-control training, inference, benchmarking, or infrastructure-heavy deployments.
Sandboxes Target Agents, Testing, and Disposable Development
Sandboxes move in the opposite direction. Instead of giving users a full physical machine, Targon is offering short-lived, isolated Linux environments that can be created quickly and discarded when no longer needed.
Targon said Sandboxes can move from provisioning to running within seconds and can be accessed through a browser terminal, graphical Linux desktop, or SSH. The product is aimed at development, testing, preview environments, and agent workloads, where users may need many temporary environments rather than one persistent server.
Sandboxes include several workflow features. Users can fork independent copies for parallel experiments or agent tasks, publish reusable templates from configured environments, and pause idle Sandboxes to stop active compute billing before resuming them later.
Those features map closely to how AI agent and software-development workflows are evolving. Agents often need isolated environments where they can run commands, test code, inspect outputs, and recover from failed attempts without affecting production systems. Developers also need repeatable environments for previews, continuous testing, experiments, and short-term collaboration.
By adding Sandboxes, Targon becomes both a GPU compute marketplace and an execution layer for software agents and rapid development loops.
The launch gives Targon a wider product range at a time when decentralized compute networks are trying to move from raw capacity claims toward more usable cloud products.
Compute buyers rarely need only one deployment model. Some want confidential execution, others need full hardware control, and others need ephemeral environments that can be created, forked, paused, and discarded. By supporting Bare Metal and Sandboxes alongside its existing platform, Targon can serve more of those demand patterns from the same subnet ecosystem.
