BroadChain News, April 23, 19:16 - Capital expenditure on AI infrastructure is undergoing an unprecedented expansion cycle. From NVIDIA GPUs to data centers, the entire industry chain is operating at high speed. However, assets like GPUs have an inherent contradiction: they generate cash flow quickly but depreciate just as rapidly. Traditional banking systems tend to be cautious towards such high-volatility, rapidly iterating assets.
USD.AI aims to address this pain point. Developed by Permian Labs, its core logic is "off-chain organization, on-chain lending" — channeling on-chain dollar liquidity into heavy asset areas such as GPUs, cabinets, and data centers. Borrowers must first place orders with OEMs, secure data center rack space, and establish SPVs. Subsequently, GPUs are installed and electronic warehouse receipts are generated. These receipts are tokenized into ERC-721 NFTs and ultimately used as collateral to obtain USDC loans.
Currently verifiable cases include: In April 2026, Crucible Capital drew a loan of $26.82 million, collateralized by 72 units totaling 576 NVIDIA B300 GPUs, deployed in Washington State, USA. The project has not fully disclosed the list of OEMs, data center operators, and borrowers, but known entities include QumulusAI, Quantum Solutions, Sharon AI, and Crucible. These companies are all computing infrastructure providers, not consumer-grade AI products.
USD.AI adopts a three-tier token structure: USDai serves as the base liquidity layer, pegged 1:1 to the US dollar and non-yielding; sUSDai is the yield layer, capturing GPU financing returns, with exits requiring asynchronous redemption and the QEV mechanism; CHIP is the governance token, allowing holders to participate in core decisions such as collateral standards, loan interest rates, and risk parameters, but without rights to protocol revenue distribution. As of April 23, USDai's circulating market cap is approximately $280 million, TVL is about $283 million, active loans are around $60 million, and revenue over the past 30 days is roughly $850,000.
On the risk front, GPU annual depreciation rates may reach 15%-20%. The project mitigates risks through low LTV, three-year amortization, the Barker valuation system, and reinsurance-supported value warranties. However, borrower cash flow depends on the actual execution of computing power contracts, and the outside world currently lacks transparent data to verify their quality. Additionally, the liquidity mismatch between long-term GPU loans and short-term redemptions, while smoothed by the QEV mechanism, cannot be fundamentally eliminated. The governance power of the CHIP token may in practice be concentrated among institutional participants, and its influence remains to be seen.