What is inference as yield?

Inference as yield, also called yield-as-inference, turns the rewards a staked crypto asset earns into AI compute instead of paying them out as more tokens. You keep the asset that produces the rewards, and the rewards pay part of your AI bill. Gatewayz runs one version of it: you delegate ADA from your own wallet and receive an inference discount instead of ADA rewards.

What does inference as yield mean?

It means the reward from a staked asset is paid in AI compute. In ordinary staking, a proof-of-stake network pays you in its own token for helping secure it; in inference as yield, that reward is redirected to cover AI requests you would make anyway.

The word "inference" is the AI side. Inference is running a trained model on new input: every chat message, code completion or agent step is one request, billed per use in tokens. Because it is billed per use, it behaves like a utility bill, which is what makes it a natural place to spend a recurring reward.

Why did inference as yield emerge?

Because two recurring flows met. Proof-of-stake holders receive rewards every reward period, and AI users now pay for inference every month. If you hold the asset and pay for AI, receiving the reward as AI compute removes a step: no selling tokens, no moving funds, no converting to pay a card bill.

It also changes what the holder is exposed to. A holder who sells rewards to pay for AI must time that sale; a holder whose rewards become a discount takes the conversion at the moment rewards are recorded. Neither removes price exposure on the asset itself, which stays in the holder's hands either way.

What are the possible designs?

There are two choices that define any inference-as-yield design: what the reward becomes, and where your asset sits while it earns. The table shows the four combinations.

Credits vs a discount. Credits behave like a gift card: a dollar amount you spend down at full price. A discount lowers the price you pay per request while a balance lasts, so you still pay something on every call and the balance is drawn at the discounted share. Either way, the value is tied to one provider's inference and is not cash.

Custodial vs delegation. In a custodial design you transfer the asset to a provider, which stakes it and passes rewards back as compute; you rely on the provider to return it. In a non-custodial design you only assign your stake, and the asset never leaves your wallet. On Cardano, delegation assigns your stake to a stake pool without transferring your ADA, and staking has no lock-up. Non-custodial vs custodial staking compares custody across chains.

How does the Gatewayz design work?

Gatewayz uses delegation and a discount. You point ADA at the Gatewayz pool from any CIP-30 wallet, link the wallet to your account by signing a message, and after each epoch Gatewayz turns its share of the pool's rewards into your discount balance.

The pool sets its margin near 100%, so the rewards that would normally be paid to delegators in ADA are instead kept by the pool as pool fees. Each day, after a finished epoch's rewards are recorded, Gatewayz splits what it actually received across linked wallets by stake and credits 95% of each wallet's share, in USD at the time of recording, to your inference discount balance. While that balance lasts, you pay 1% of each request's list price and the other 99% is drawn from the balance.

The full mechanism, including how a finished epoch is split across linked wallets, is on How it works.

Who does inference as yield suit?

It suits holders who already pay for AI every month and would rather not sell to do it. Three groups fit best:

• Teams and treasuries that hold ADA and pay a recurring AI bill. See For treasuries.
• Builders who use OpenAI-compatible tools and can fund their own usage from stake they hold anyway. See For builders.
• Agent operators whose agents make many small requests from one API key. See For agents.

Who does it not suit?

It does not suit anyone who wants their rewards as tokens. With the Gatewayz pool you give up ADA staking rewards while delegated, and the discount cannot be withdrawn, transferred or exchanged. If you want to accumulate ADA, delegate to a pool that pays rewards out.

It also does not suit people who rarely use AI, since a balance you do not spend does nothing for you, or anyone in the United States or Canada: the program is not available to US or Canadian persons. And it is not live on mainnet yet.

Frequently asked questions

Is inference as yield the same as yield-as-inference?

Yes. Both names describe the same idea: protocol rewards from a staked asset are turned into AI compute instead of being paid out as tokens. Gatewayz uses the term yield-as-inference for its program.

Is an inference discount the same as cash?

No. In the Gatewayz design the discount balance is non-transferable, works only for inference on Gatewayz, and cannot be withdrawn, exchanged or cashed out. It lowers the price of requests you make; it is not money you can move.

Do I have to give my crypto to a provider?

Not in a delegation design. On Cardano, delegation assigns your stake to a pool with a certificate on your stake key, and your ADA stays in your wallet and spendable. Custodial designs, where you deposit the asset, do require trusting the provider to return it.

Why a discount instead of credits?

A discount means you pay a small part of every request and the balance covers the rest, so the balance lasts across many requests and you always see the real list price. At Gatewayz you pay 1% of list price and the balance covers 99% while it lasts.

Can I use inference as yield with assets other than ADA?

Not yet at Gatewayz. ADA is the only live asset; other proof-of-stake assets and stablecoins are listed on the assets page so holders can tell Gatewayz which should come next.

How much AI does it pay for?

It depends on how much you stake, how many blocks the pool produces and the asset's price when rewards are recorded. There is no promised rate; the Gatewayz calculator gives a range for your own amount with the launch limit applied.

Sources

• Cardano.org Glossary: Delegation, as of 2026-10-27
• Cardano.org: Stake your ada on Cardano, as of 2026-10-27
• CIP-0030: Cardano dApp-Wallet Web Bridge, as of 2026-10-27