How Does USD.AI Work?

RWA|Risk D|5 mechanisms|5 interactions

USD.AI (USDai) is a synthetic dollar backed by loans against AI infrastructure, specifically NVIDIA GPUs. Built by Permian Labs, it uses a novel system called CALIBER to tokenize physical GPU hardware as NFTs, then issues loans against this compute collateral. The staked version (sUSDai) earns 13-17% APY from GPU operator loan repayments. With $658M TVL and backing from Framework Ventures, Dragonfly, and Coinbase Ventures, it has attracted significant capital. However, the entire model depends on GPU hardware maintaining value and AI compute demand staying strong — both assumptions that carry significant risk given the rapid pace of hardware depreciation and the cyclical nature of technology spending.

TVL

$288M

Sector

RWA

Risk Grade

D

Value Grade

C

Core Mechanisms

Lending/Asset-Backed

Novel

Loans collateralized by NVIDIA GPU hardware operated by AI compute providers

First protocol to use physical compute infrastructure as DeFi loan collateral. GPUs are valued and tracked via CALIBER registry.

Tokenization/Hardware-NFT

Novel

CALIBER registry tokenizes GPU hardware as NFTs with offchain+onchain verification

Novel tokenization of physical hardware assets. Each GPU rack becomes an NFT representing ownership and operational status.

Yield/Loan-Repayment

Novel

sUSDai yield derived from GPU operator loan interest payments (13-17% APY)

Yield comes from real-world loan repayments by compute operators, not on-chain activity. Unusual for DeFi stablecoins.

Settlement/Fiat-Integration

Novel

PayPal PYUSD integration for fiat-crypto loan settlement

Partnership with PayPal for PYUSD settlement adds fiat on/off ramp but also introduces PayPal counterparty dependency.

Valuation/Compute-Pricing

GPU hardware valuation model for collateral assessment and loan-to-value ratios

Proprietary valuation model for NVIDIA GPUs. Must account for rapid depreciation curves and compute market pricing.

How the Pieces Interact

Lending/Asset-BackedValuation/Compute-PricingCritical

GPU depreciation could outpace loan repayment schedules, leaving loans undercollateralized with no trustless liquidation mechanism for physical hardware

Tokenization/Hardware-NFTLending/Asset-BackedHigh

NFT representations of hardware may diverge from actual physical asset status — equipment could be damaged, moved, or double-pledged without on-chain detection

Yield/Loan-RepaymentSettlement/Fiat-IntegrationMedium

PayPal settlement dependency means a PYUSD depeg or PayPal service interruption could block loan repayments, halting sUSDai yield

Lending/Asset-BackedYield/Loan-RepaymentHigh

Mass GPU operator defaults during an AI market downturn could simultaneously crash collateral values and eliminate yield, creating a death spiral

Tokenization/Hardware-NFTValuation/Compute-PricingMedium

Hardware NFT valuations may lag real market prices for GPUs, especially during rapid depreciation from new GPU generation launches

What Could Go Wrong

  1. GPU hardware depreciates rapidly — NVIDIA GPUs lose 50%+ value within 2-3 years as new generations launch
  2. AI compute demand is cyclical — an AI winter could crash collateral values and loan repayment capacity
  3. Physical asset custody and verification is fundamentally harder than on-chain collateral — no trustless liquidation
  4. Novel collateral type with zero liquidation precedent in DeFi or TradFi at this scale
  5. Concentration in NVIDIA hardware creates single-vendor dependency risk

AI Winter Collateral Collapse

Moderate

Trigger: AI spending contracts sharply as enterprises cut compute budgets, crashing GPU demand and operator revenues

  1. 1.AI market downturn causes GPU compute demand to drop 40%+, crashing GPU resale values and operator revenue Loan collateral (GPUs) depreciates faster than repayment schedule; LTV ratios breach safety thresholds
  2. 2.GPU operators default on loans as compute revenue can no longer cover interest payments sUSDai yield drops to zero; protocol holds depreciating hardware it cannot efficiently liquidate
  3. 3.Physical hardware liquidation is slow (weeks/months) and illiquid, unlike on-chain collateral USDai backing falls well below 1:1; redemptions cannot be met, triggering depeg and confidence crisis

Risk Profile at a Glance

Mechanism Novelty12/15
Interaction Severity14/20
Oracle Surface5/10
Documentation Gaps7/10
Track Record12/15
Scale Exposure5/10
Regulatory Risk8/10
Vitality Risk5/10
D

Overall: D (68/100)

Lower score = safer

More on USD.AI

Related RWA Explainers