Skip to main content
Calling a model today usually means handing a payment method to a cloud provider and letting it meter your usage. Staking for Inference removes that dependency: you stake NEAR, and the staking rewards — not the staked NEAR itself — pay for private inference on NEAR AI Cloud. You keep the NEAR. Unstake, and the credits stop.

​
How it works

Confidential inference converts the yield your stake earns, rather than the stake itself, into compute credits. Your monthly budget is a function of three things:
  • How much NEAR you stake
  • The price of NEAR
  • The current staking APY — the same yield your stake would earn anyway, redirected into compute
Credits accrue continuously, on a per-second basis, as a function of your stake size — there’s no waiting for a monthly refresh. One credit is worth one US dollar of usage on NEAR AI Cloud, the same unit your bill is already denominated in; unused credits carry forward.
Because the reward rate is network-wide, not a fixed setting, your covered usage drifts with both the NEAR price and how much of the total supply is staked. Try the staking calculator to see current numbers.

​
Why the reward rate moves

NEAR issues a fixed amount of new token each year and splits it among everyone staking. Your share of that depends on how much of the supply is staked network-wide: when less is staked, the same rewards divide among fewer tokens and the rate rises; when more is staked, it falls. Validator commission, typically 1–10%, comes out of that yield before it reaches you.

​
What you keep

Two things follow from paying with staking yield instead of a metered card:

Principal stays yours

Staked NEAR remains withdrawable. You retain ownership of the underlying stake the entire time it’s generating credits — running inference is a recoverable position, not a sunk bill.

Reversible, any time

Unstake and you walk away with your original NEAR intact. The credit allowance simply stops accruing — nothing is spent or forfeited.

​
Confidentiality guarantees

Staking for Inference pays for the same private inference NEAR AI Cloud always provides — it doesn’t change the privacy model, only how you pay for it.
  • Open-source models (Qwen, DeepSeek, GLM) run inside a Trusted Execution Environment — a hardware-isolated enclave on the GPU. Your prompt and the model’s output are sealed off from the host operating system, the GPU operator, and NEAR AI itself. Each request returns a hardware-signed attestation in under 30 seconds, so this is a guarantee you can check, not a promise you have to trust.
  • Frontier models (from providers like Anthropic, OpenAI, and Google) route through a secure gateway that itself runs inside a TEE. The provider sees a prompt arrive from the gateway and nothing else — it cannot tie the request back to you. An optional PII redaction step strips identifying details before the request leaves the enclave.

​
Managing your stake

The entire lifecycle lives in your NEAR wallet. Stake more for a larger credit budget, or unstake to reclaim your NEAR — every change is a signed wallet transaction, with no separate account or credit card involved.

​
Next steps

Staking Calculator

Work out how much NEAR to stake for your usage, or what a given stake gets you.

Private Inference

See how NEAR AI Cloud’s TEE architecture protects your data end to end.