Tenzro Network is an economy for distributed compute and decentralized AI, open to machines of every size, from home nodes to data centers. It works alongside centralized data centers, whose operators are supply and customers of the same rails. This post is for anyone who wants to run a machine on it: which roles exist, what each needs and how admission works while the network is in preview.
The roles
An operator account has one hardware identity and can take several roles. Each role is bonded separately from the same account, and the amounts are governed parameters. The roles that matter most to an operator:
| Role | What you provide | Node role |
|---|---|---|
| AI provider | Inference for open models | ai |
| Compute provider | Rented GPU and CPU capacity | compute |
| Data and storage provider | Content-addressed storage | storage |
| TEE provider | Confidential execution with attestation | tee |
| Validator | Orders and finalises blocks | validator |
| RPC provider | Public JSON-RPC and API access under your access policy | any serving node, bonded as rpc |
The full list, with the commands, is in Operators and roles. Payment follows paid, verified use: nothing is paid for being listed, for holding hardware or for having signed up. Whether a provider is used depends on demand and on what the machine can serve.
Providers: compute, AI and data
Anyone who bonds for a role can serve it, and there is no list of providers to be added to. The requirements are documented in Join the network as a provider:
- a release archive from Supported hardware, checked with
tenzro-node hardware; - a hardware key: a TPM 2.0, a Secure Enclave, or another machine's TPM through a key holder;
- outbound UDP 9001 and TCP 443, with no inbound port required;
- TNZO for the role bonds and gas in the operator account;
- for models, acceptance of the licence terms the node asks for.
The CPU build serves models of 6 GiB of weights or less. Larger models need the GPU build on a machine with a supported GPU. A small or mid-size operator can start with the machines it already runs and add roles later.
An operator can be both an AI provider and a data provider, as far as the code supports it. The protocol has no separate data-provider role. What an operator can do with data today is hold it for others as a storage provider, paid hour by hour on possession proofs, publish registry entries for datasets and services, and run its own application beside the node. A dataset marketplace with its own settlement is not part of the launched network. Roles on one account share the machine's disk, memory and bandwidth.
Validators
Any TPM 2.0 machine can run a validator node. During the preview voting validators are admitted by governance approval: while the governed switch validator.admission_approval is 1, as in the launch genesis, an approve_validators proposal must have named the account first. A validator registers with a stake set by a governed parameter. See Validator lifecycle.
A validator's consensus key lives in a TPM in its own machine, or is reached through a key holder. A validator does not need a GPU.
RPC operators
Any serving node can offer public RPC. Bond with the rpc role, publish the endpoint on chain, and set who may call you: open, paid per call, or the DIDs and rate tiers of your subscribers. Serving RPC publicly needs your own TLS reverse proxy. See Run an RPC provider.
Sponsorship and grants
Sponsorship and grants are described in the node repository's docs/SPONSORSHIP-AND-GRANTS.md and in Sponsorship and grants. The facts as written there:
- The treasury pays two things from TNZO it already holds: sponsorship of approved validators and providers, and grants to approved developers, contributors and community members. Nothing is issued.
- Every disbursement is one governance proposal, except two unattended releases that pay verified work and carry their own caps.
- Approval records an operator with the criteria it was approved on, confers no vote, no reward and no class, and can be revoked, which stops later payments.
- Operator running costs are paid only against hours the challenge protocol judged the operator available.
- All caps start at zero, so nothing flows until governance sets them. An approved operator is paid nothing until then.
- A new operator cannot file the approval itself: the proposal needs a proposer holding 1% of the bonded stake.
tenzro sponsor check,applyandstatusread the criteria, write an application and show the record. A Secure Enclave has no TPM endorsement certificate and a Jetson's firmware TPM has no recognised root, so neither can be approved today.
TEE-backed confidential providers
A provider with Intel TDX, AMD SEV-SNP, AWS Nitro, or an NVIDIA GPU in confidential-computing mode inside a confidential VM can bond the tee role. There is no separate security-provider role. A TPM and Secure Boot alone do not satisfy a confidential-compute requirement.
What an attestation shows: which software ran on which platform at which security level, verified against the vendor's certificate chain, with a fresh nonce so old evidence cannot be replayed. What it does not show: that the software is free of bugs, that the platform vendor is out of the trust chain, or that a flaw later found in a platform did not apply. A TEE narrows who can read data while it is processed. It reduces trust and does not remove it. See Trusted execution environments.
What is not claimed
The network is in preview and has not yet carried production use. This post gives no earnings or return figures, because there are none to give: what a provider is paid depends on paid demand and on the machine's utilisation. That a small board joins and is paid, that one request crosses tiers and settles, and that the largest models run across machines are designed but not yet measured on the launched network.
