sUSDai

sUSDai is the yield-bearing token of the USD.AI protocol.

sUSDai is not a stablecoin. It is not instantly redeemable at par value. It represents a claim on underlying assets that include illiquid GPU-collateralized loans, and redemptions are subject to epoch-based timing constraints.

Token Mechanics

USDai is staked to mint sUSDai, which captures yield from the underlying protocol activities in the exchange ratio between USDai and sUSDai . This appreciation mechanism means that sUSDai is not pegged 1:1 to any other asset.

Conversion Rate Formula:

The amount of sUSDai received from staking is determined by the current conversion ratio.

sUSDai received = USDai deposited × (Total sUSDai Supply / Total USDai in Vault)

The amount of USDai received from unstaking:

USDai received = sUSDai redeemed × (Total USDai in Vault / Total sUSDai Supply)

At protocol launch, the conversion rate was 1:1. Staking 100 USDai minted 100 sUSDai. As yield accumulates in the vault, the conversion rate shifts. If the vault holds 110 USDai backing 100 sUSDai, the rate becomes 1.10 USDai per sUSDai. Staking an additional 100 USDai would mint yield approximately 90.9 sUSDai.

Exchange Ratio

sUSDai employs a rebasing-by-price model rather than a rebasing-by-supply model. The number of sUSDai tokens a holder owns does not change; instead, each sUSDai token becomes redeemable for an increasing amount of USDai as yield accrues.

Yield Generation

Yield accrues continuously as:

• Interest payments are received from active AI infrastructure loans

• Idle capital generates returns from Treasury Bill allocations

GPU-Collateralized Loans

The USD.AI protocol generates yield primarily through asset backed, secured loans to AI infrastructure operators, typically data center operators acquiring GPU hardware. These loans are structured as follows:

• Collateral: Physical GPU hardware (NVIDIA B300, B200, H200, RTX Pro 6000, and other enterprise-grade high performance computing hardware)

• Borrowers: Data center operators, cloud compute providers, and AI infrastructure companies

• Loan Purpose: Hardware acquisition financing, analogous to equipment financing or secured asset-based lending

• Interest Rates: Variable based on LTV and offtake type, ranging from 7-15% APR

The loans function similarly to traditional equipment financing or mortgage structures, borrowers make periodic payments covering principal and interest, with the underlying hardware serving as collateral.

Secondary Source: Treasury Bill Floor

Capital not actively deployed in GPU loans is allocated to short-term U.S. Treasury Bills. This creates a yield floor for idle capital, minimizing any cash drag caused by timing gaps between deposit / staking events and loan originations.

The T-Bill allocation serves two purposes:

1. Yield continuity: Depositors earn baseline returns regardless of loan deployment timing

2. Liquidity buffer: T-Bill positions can be liquidated to service redemptions without forcing premature loan terminations

sUSDai Redemption Mechanics

Redemptions operate on a global 30-day epoch cycle. This is a protocol-wide timer, not a per-user countdown. Current withdrawal mechanics use a FIFO queue, with plans to transition to QEV in the future.

Timeline within each epoch:

Period	Action
Day 1–29	Users may queue unstake requests at any time
Day 29 (cutoff)	Queue closes; no new requests accepted for current epoch
Day 30 (epoch close)	Protocol processes queued redemptions from available cash, USDai returned

Queue Processing: FIFO with Cash Constraints

At epoch close, redemptions are processed as follows:

1. Available USDai is calculated: This is USDai held in the vault that is not actively deployed in loans or reserved for imminent loan funding

2. Queue is processed FIFO: Redemption requests are serviced in the order they were submitted (first-in, first-out)

3. Full USDai release: The protocol releases all available free USDai to service the queue. There is no throttling or partial release mechanism

4. Scenario A—Sufficient USDai: If available USDai covers all queued redemptions, every request in the queue is fulfilled in full

5. Scenario B—Insufficient USDai: If available USDai is insufficient to cover all queued redemptions:

   1. Requests are fulfilled FIFO until USDai is exhausted

   2. Partially fulfilled requests receive their partial amount; the remainder carries forward

   3. Unfulfilled requests carry to the next 30-day epoch

   4. No priority is given except queue position

Critical Constraints

The protocol does not prematurely terminate or liquidate active GPU loans to satisfy redemptions. Loans run to maturity or scheduled paydown. This means:

• During periods of high utilization (most capital deployed in loans), available cash may be limited

• Redemption queues may extend across multiple epochs during high-utilization periods

• Users cannot force immediate liquidity from the protocol

QEV

A Queued Redemption Engine with auction-based priority (see: QEV) is planned but not yet implemented as of April 2026. QEV will allow users to bid for priority in the redemption queue, providing a mechanism for time-sensitive exits at a cost. Implementation timeline and specifications are not finalized.