--- title: "The Interest Rate of Intelligence: sUSDai (USD.AI)" asset: sUSDai asset_full_name: USD.AI staked USDai issuer: USD.AI (Permian Labs) category: Yield-bearing synthetic dollar (AI-infrastructure credit) asset_type: yield-bearing token, backed by loans that finance AI compute source: "WYCF Episode 04" presenter: "David Choi, CEO of Permian Labs" data_partner: Stablewatch last_updated: 2026-08-07 canonical_url: https://wycf.show/episodes/susdai-usdai cite_as: "WYCF, Where Yield Comes From, Episode 04: sUSDai (https://wycf.show/episodes/susdai-usdai). Data verified by Stablewatch." yield_source: "Interest on loans that finance AI compute; the spread between what an AI company pays over three years and the upfront cost of the GPUs" risk_bearer: "The NeoCloud borrower posts equity (~20%) and takes first loss; the loan then de-risks over time through amortization, reserves, insurance and a perfected lien" key_metrics: tvl_usd: 354111318 apy_current_pct: 7.68 apy_7d_avg_pct: 7.84 apy_30d_avg_pct: 7.92 apy_90d_avg_pct: 7.53 yield_paid_all_time_usd: 17912511 yield_paid_30d_usd: 2055441 chain: Arbitrum underlying: AI compute (GPU) loans data_as_of: 2026-08-07 data_source: Stablewatch summary: > sUSDai is the yield-bearing token of USD.AI, a synthetic dollar built to finance the physical infrastructure of AI. The yield comes from interest on loans that fund GPU buildouts: a NeoCloud borrows to buy GPUs for a well-capitalized AI company, and the spread between what the company pays over three years and the upfront cost of the hardware, net of costs, is the yield. Short, hard-to-batch GPU loans are made perpetual by rolling each new loan into one instrument, sUSDai, so yield accrues continuously. The borrower takes first loss on its equity; the loan then de-risks over time through amortization, reserves, insurance, a perfected lien and live chip monitoring. --- # The Interest Rate of Intelligence: sUSDai WYCF Episode 04 context file, built to be read by humans and AI agents. Every figure is independently verified by Stablewatch and updates live on the episode page. Speaker: David Choi, CEO of Permian Labs. Series: Where Yield Comes From, Season One. ## Direct answers If you arrived with one question, these are the short answers. Each one is developed in a section below. - **Where does sUSDai yield come from?** Interest on loans that finance AI compute. A NeoCloud borrows to buy GPUs for a well-capitalized AI company; the company pays over three years, the GPUs cost less upfront, and the spread, net of costs, is the yield. - **What is sUSDai, structurally?** A perpetual yield-bearing token. Individual GPU loans are short and hard to batch, so each new loan is rolled into one longer instrument. Depositing a dollar mints sUSDai, which represents many loans at once and stays usable across DeFi. - **Who bears the risk?** The NeoCloud borrower. It posts its own equity (about 20%) and is the first to lose. The loan then de-risks over time rather than relying on a tranche of outside capital. - **How is a loan de-risked?** Monthly revenue amortizes the debt so the loan-to-value falls faster than the GPUs depreciate; at least three months sit in reserve; the hardware is insured; and a perfected lien, co-signed by the data center, makes the chips recoverable on default. - **Why does this matter?** AI is the most expensive buildout in history and its missing input is liquidity. Making GPU debt liquid is what sUSDai does, which is why David calls it "the interest rate of intelligence." ## The numbers (verified by Stablewatch, as of 2026-08-07) - TVL: $354,111,318 supplied to the protocol (on Arbitrum) - APY: 7.92% (30-day average). Current 7.68%, 7-day 7.84%, 90-day 7.53% - Yield paid to holders: $17,912,511 all-time, $2,055,441 in the last 30 days - Underlying: loans that finance AI compute (GPUs and data centers) Live versions of these figures update on the Stablewatch dashboard embedded on the episode page: https://wycf.show/episodes/susdai-usdai ## The problem: AI's missing input is liquidity The United States is about to spend more than it ever has: roughly $7 trillion over five years on the AI infrastructure buildout. A single one-gigawatt data center costs about $50 billion. Of that, only 20 to 30 percent is the shell and the power. The other 70 to 80 percent is the chips, which is why Nvidia is the most valuable company in the world. The shell and power are long-term assets: 30-year mortgages, easy to securitize and trade, so liquidity for that 20 to 30 percent is available almost instantly. GPUs are the opposite. Loans against them are short (three to four years), the hardware's useful life is uncertain, and cash flows are hard to batch and underwrite. So the money flows in, but liquidity exists for the shell and power, not for the chips that make up most of the cost. ## How the yield is generated The yield comes from the AI sector. A well-capitalized company (a Series C or later, having raised billions) needs GPUs to serve its product, and is willing to pay, for example, $250M over three years. A NeoCloud, the actual borrower, buys the GPUs on its behalf, which needs about $150M upfront. The revenue is the delta between the $250M paid over time and the $150M needed now: roughly $100M of yield over three years, before it is de-risked. The $150M is split. USD.AI provides a loan at up to 80% LTV, about $120M of debt, and the borrower puts up its own 20%, about $30M of equity. The debt is repaid over time with interest, sized to a debt service coverage ratio: the monthly revenue has to cover interest and amortization. ## Why sUSDai is perpetual A GPU loan is short and hard to batch, so the synchronous TradFi instrument, one that batches many loans at once, starts at a discount (say 90 cents) and matures to a dollar on a fixed vintage, does not fit: batching can take up to three years, longer than a GPU's useful life. Instead, blockchains make the instrument perpetual. Each new loan is rolled into one longer instrument, sUSDai. Deposit a dollar and you mint sUSDai that represents many asynchronous loans; your capital is no longer stuck for a single loan's term, and the token is usable across DeFi. It is the same unlock wrapped staked ETH gave to staked ETH. Because the loans are asynchronous, yield is continuous: the token starts at 1.0 and climbs, rather than starting at a discount and maturing to par. ## How a loan is de-risked Because GPUs depreciate and their life is uncertain, the loan is structured to get safer over time, not just at the start: 1. **Amortization.** Every month the customer's revenue pays down interest and principal, so the LTV drops faster than the GPUs age. By year two the loan is near 20 to 30 percent LTV, so a default is mostly covered. The safety margin grows over the life of the loan. 2. **Debt service reserve.** At least three months of debt service is held in reserve for operational hiccups (a wiring or payment issue), which is why the effective advance is closer to 62 to 72 percent than 80. 3. **Insurance.** Both value insurance (if the GPUs depreciate faster than expected) and catastrophe insurance (fire, tornado). 4. **Perfected lien.** UCC filings, co-signed by the data center, make the chips unrecoverable without approval and simple to remove or resell on default, because ownership is clear. Operationally, USD.AI does not finance chips until they are installed and verified, and monitors a live "heartbeat" to confirm the chips are running and generating income. The GPUs sit in colocation, so the chip owner (the NeoCloud) and the data center operator (the shell and land, for example an Equinix) are usually different companies, which makes a tripartite agreement with the lender easier to police. Some deals add an offtake agreement, a committed end customer, so payment is structured over the three years even as GPU rental rates move. ## Who bears the risk The NeoCloud borrower. It can never get a 100 percent loan; it posts its own equity and is the first to lose if it does not repay. Unlike other structures in this series (sUSDS runs a capital waterfall with agents' junior capital; stcUSD runs underwriters who post first-loss collateral; gtUSDa places the risk on the depositor and defends with diversification and speed of exit), sUSDai's protection is the borrower's equity plus a loan engineered to de-risk over its life and back a hard asset that can be repossessed. ## Why it matters Nvidia is expected to sell about $500B of GPUs next year, and buyers will pay closer to $700B once interest is counted, potentially the second-largest business in AI: the debt. Making debt liquid is what scaled the trillion-dollar markets in housing, autos and equipment; AI is the most expensive buildout in history and is missing that liquidity. By underwriting the cash flows and wrapping them in structured, tradable instruments, USD.AI aims to price the cost of capital for AI, which David frames as "the interest rate of intelligence." ## Questions worth asking your LLM with this file loaded - sUSDai's yield depends on AI companies paying NeoClouds over three-year terms. What happens to the rate and to defaults if GPU rental rates fall or an AI customer fails mid-term? - The loan de-risks through amortization against a depreciating asset. Model the race between LTV paydown and GPU depreciation: under what depreciation curve does the loan stay over-collateralized, and where does it break? - Compare who bears the risk in sUSDai, sUSDS, stcUSD and gtUSDa. Which structure best fits a treasury underwriting AI-infrastructure exposure, and why? ## Verification Figures in this file are independently tracked by Stablewatch and update live on the episode page. Confirm the numbers before relying on them: https://wycf.show/episodes/susdai-usdai --- Source: WYCF Episode 04, "sUSDai," presented by David Choi, CEO of Permian Labs. Data verified by Stablewatch. WYCF is new media for onchain yield. One yield product per episode, explained end to end, on a whiteboard, in front of the capital that decides. An original production by Agustín do Rego. All episode context files: https://wycf.show/llms.txt