Open Etherscan, look up the address 0x37305b1cd40574e4c5ce33f8e8306be057fd7341 and check its USDC balance. On 28 October 2026 at about 09:45 UTC it held roughly 3.8 billion USDC. That wallet is the reserve of Sky’s LitePSM, the module that swaps the DAI stablecoin and its successor USDS for USDC at exactly one dollar. On the same day DAI and USDS together had about 11.4 billion tokens in circulation, according to DefiLlama. So a pile of Circle-issued dollars equal to about a third of that supply sits behind a coin that most people file under “crypto-backed”.
That one number is a good way into collateralised debt position stablecoins, or CDP stablecoins: dollars that are not deposited at a bank but created by a loan contract. Someone locks crypto, borrows newly minted coins against it, and the coins disappear again when the loan is repaid. The idea goes back to MakerDAO’s first Dai in December 2017. In 2026 it runs in four main designs: Sky (the former MakerDAO, with DAI and USDS), Liquity (LUSD and BOLD), Aave’s GHO and Curve’s crvUSD. This lesson takes them apart lever by lever, with the parameters read on-chain on 28 October 2026.
A short reminder, then the numbers: where CDP stablecoins stand
A stablecoin is a token that aims to stay worth one unit of a currency, usually one dollar; our guide to what stablecoins are and how they hold their peg covers the families, and USDT versus USDC explains the two big bank-reserve coins. This lesson assumes you know those basics and that you have read the previous lesson of the path, on composability and DeFi money legos, because CDP stablecoins are the most stacked lego of all: they are minted from deposits in one protocol and spent in others.
In size, CDP coins are a small corner of the market. DefiLlama counted about 302.5 billion dollars of dollar-pegged stablecoins on 28 October 2026, of which roughly 274.7 billion were backed by fiat reserves. The ones covered here look like this:
| Stablecoin (issuer) | Design | Circulating supply, 28 Oct 2026 |
|---|---|---|
| USDT (Tether) | Fiat reserves (for comparison) | 181.18 billion |
| USDC (Circle) | Fiat reserves (for comparison) | 71.30 billion |
| USDS (Sky) | CDP vaults plus PSM | 6.667 billion |
| DAI (Sky) | CDP vaults plus PSM | 4.703 billion |
| GHO (Aave) | Borrowed from a lending pool | 0.698 billion |
| crvUSD (Curve) | CDP with soft liquidation | 0.222 billion |
| BOLD (Liquity v2) | CDP with user-set rates | 0.038 billion |
| LUSD (Liquity v1) | CDP with no interest | 0.026 billion |
Source: DefiLlama stablecoins data, all chains, retrieved 28 October 2026 between 09:45 and 10:30 UTC. DefiLlama also files USDe under “crypto-backed”, at about 4.5 billion, but Ethena’s synthetic dollar works with derivatives positions, not loans, and is the subject of the next lesson.
The five levers every CDP stablecoin has to set
Strip away the branding and every system in this lesson answers the same five questions. Comparing them lever by lever is faster than learning each protocol from scratch.
The first lever is the collateral ratio: how many dollars of collateral must sit behind each dollar of debt. Sky’s ETH-A vaults need 145%, Liquity v2’s ETH branch 110%. A lower ratio lets borrowers mint more per ether but leaves a thinner cushion when prices fall.
The second is the price of borrowing. It can be a fee voted by token holders (Sky’s stability fee, Aave’s GHO rate), a formula tied to the peg (crvUSD), a rate each borrower picks (BOLD), or zero with a one-off fee (LUSD).
The third is liquidation, the rule that sells collateral before a loan becomes worth more than its backing. Sky runs auctions, Liquity hands collateral to a blockchain that runs exactly as written when it is called, holding and moving funds according to its code, without an intermediary able to change the outcome.">smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool of depositors, Curve converts it gradually inside an automated market maker. Our glossary entry on liquidation gives the general idea; the details here are where the designs differ most.
The fourth is a door at one dollar: some way to swap the coin for close to a dollar of something else when it trades below the peg, or to create more of it when it trades above. Without that door, a stablecoin is only stable by hope.
The fifth is a savings rate, a yield paid to holders who park the coin in a contract. It raises demand for the coin and props up its price, but it has to be paid from somewhere, usually the interest borrowers pay.
Sky and the DAI stablecoin: from MakerDAO vaults to USDS
MakerDAO launched single-collateral Dai in December 2017 and moved to multi-collateral DAI in November 2019. On 27 August 2024 the project took the name Sky, and on 18 September 2024 it added a second coin, USDS, and a new governance token, SKY. Both stablecoins come out of the same machine, so what follows applies to each. The Sky project page covers the organisation; here we look at the contracts.
How a Sky vault mints DAI or USDS, step by step
A vault is a position in a contract called the Vat. Each collateral type has its own risk settings, known internally as an “ilk”: ETH-A, ETH-C, WSTETH-A and so on. When you open a vault you lock collateral, and you draw debt, which the system creates as fresh DAI or USDS in your wallet. The debt is stored as a normalised amount multiplied by a rate accumulator that grows every second at the stability fee. Repay the debt plus the accrued fee and the coins you return are burned; your collateral is released.
Each ilk also carries a debt ceiling, the total that all vaults of that type may mint. On 28 October 2026 ETH-A had about 180.5 million of debt against a ceiling of 329 million. These are the main parameters we read from the contracts at Ethereum block 26,290,413 (the liquidation ratios and penalties were read again at block 26,290,502):
| Vault type | Stability fee (annual) | Minimum collateral ratio | Liquidation penalty | Debt outstanding | Debt ceiling |
|---|---|---|---|---|---|
| ETH-A | 10.50% | 145% | 13% | about 180.5 million | 329 million |
| ETH-C | 10.25% | 170% | 13% | about 177.5 million | 284 million |
| ETH-B | 11.00% | 130% | 13% | about 6.0 million | 29 million |
| WSTETH-A | 11.50% | 150% | 13% | about 18.1 million | 60 million |
| WSTETH-B | 11.25% | 175% | 13% | about 5.3 million | 50 million |
| WBTC-A | 15.50% | 150% | 0% | about 0.6 million | 0 (closed to new debt) |
| LSEV2-SKY-A (SKY collateral) | 7.41% | 120% | 13% | about 161.0 million | 185 million |
Source: Sky contracts Jug, Vat, Spotter and Dog on Ethereum, read on 28 October 2026 at about 09:45 UTC. The annual rate is the per-second rate raised to the power 31,536,000, minus one, the conversion given in Sky’s rates overview. You can repeat the reading on Etherscan with the “Read contract” tab.
A worked example, with a round price that is an illustration and not a market quote. Say ether trades at 3,000 dollars and you lock 10 ETH, worth 30,000 dollars, in an ETH-A vault. At a 145% minimum ratio, the most you could draw is about 20,690 DAI. A prudent borrower takes 15,000, a 200% ratio. The vault becomes liquidatable if ether falls to 15,000 x 1.45 / 10 = 2,175 dollars.
Now leave the vault untouched for a year. At 10.5% the debt grows to about 16,575 DAI, and the liquidation price climbs to about 2,403 dollars without any change in the market. Interest eats into the cushion. Borrowers who forget the stability fee are the ones surprised by a liquidation at a price they thought was safe.
Stability fee versus savings rate: DSR at 1.25%, SSR at 3.80%
Sky pays two separate savings rates. The Dai Savings Rate (DSR), held in a contract called the Pot, applies to DAI. The Sky Savings Rate (SSR) applies to USDS deposited in sUSDS, a vault token built on the ERC-4626 standard. On 28 October 2026 the DSR was 1.25% and the SSR 3.80% a year, read on-chain and matching DefiLlama’s yields data. A later lesson on yield-bearing stablecoins looks at sUSDS and sDAI as products; here what matters is where the money comes from.
It comes from vault borrowers and from the protocol’s other income. Sky’s documentation is plain about the order of things: “The DSR is set by Sky Governance, and will typically be less than the base stability fee to remain sustainable” (Sky docs, Pot). With stability fees of 7.41% to 15.5% on the vaults above, the gap was wide on 28 October 2026. The savings rates are a governance decision, not a market price, and they can be cut by vote.
The savings token has become the main home of USDS. sUSDS held about 4.67 billion USDS on 28 October 2026, close to 70% of the 6.68 billion USDS on Ethereum. Most USDS on Ethereum, in other words, is parked to earn the rate rather than used for payments.
Liquidation 2.0: Dutch auctions with a 13% penalty
When a vault’s collateral ratio drops below the minimum, anyone can call a function on the Dog contract that seizes the collateral and starts an auction. The debt to cover is the vault’s debt plus the liquidation penalty: 13% on every vault type in the table except WBTC-A, whose penalty was set to zero. In the example above, a vault with 16,575 DAI of debt would put 16,575 x 1.13, about 18,730 dollars of value, up for sale. Whatever collateral remains after the auction goes back to the owner.
The auction is a Dutch auction: it opens at a price above the market and falls with time until a bidder takes some or all of the lot. Settlement is instant, which allows flash lending: a bidder with no DAI can route the collateral into another protocol and pay within the same transaction, as described in Sky’s collateral liquidation documentation. A global cap, the Dog’s “hole”, limits how much debt can be under auction at once.
This design replaced the English auctions that failed on Black Thursday, 12 March 2020, when congested blocks let some bidders win collateral for zero DAI; our history of DeFi tells that day in detail. The DAI stablecoin itself stood at 1.0585 dollars on 13 March 2020 in DefiLlama’s daily price data, above its peg, as borrowers scrambled to buy DAI to repay debt.
The PSM and LitePSM: a dollar door made of USDC
The Peg Stability Module solves the fourth lever in the bluntest way possible. You hand it USDC and get the same amount of DAI or USDS; you hand back DAI or USDS and get USDC. Sky’s current version, the LitePSM, does this “with no slippage” according to its documentation, and on 28 October 2026 both of its fees (tin and tout) read zero on-chain. If DAI trades at 0.99 on an exchange, a trader buys it, redeems it for USDC at 1.00 and pockets the difference, pulling the price back up.
The cost of that door is the 3.8 billion USDC in the pocket address. Those dollars sit in a Circle token that Circle can freeze at the contract level, a power covered in the USDT and USDC comparison linked above. It also links DAI to USDC’s own peg. In March 2023, after USDC fell below a dollar during the Silicon Valley Bank failure, DAI followed it: DefiLlama’s daily data show a DAI low of 0.9669 on 12 March 2023. A CDP stablecoin that holds a large stock of a bank-backed coin inherits part of that coin’s risk.
The pocket is the module’s inventory, not the whole backing of DAI and USDS: vaults, the SKY-backed Lockstake loans and Sky’s other allocations sit alongside it. Still, the ratio is worth checking yourself from time to time, because it moves with demand.
DAI versus USDS: one machine, two coins
USDS is not a separate system. A converter swaps DAI for USDS and back at 1:1, with no fee, and Sky’s USDS documentation says “Fees cannot be enabled on this route in the future.” The differences are in the wrapping. USDS gets the higher savings rate (3.80% against 1.25% on 28 October 2026), and the USDS token contract is upgradeable through a proxy pattern, according to the same documentation, which means governance can change its code. Someone who wants a token whose code governance cannot swap has a reason to stay in DAI; someone after the higher rate has a reason to convert.
By supply, USDS has overtaken DAI: on Ethereum we read 6.678 billion USDS against 4.503 billion DAI on 28 October 2026. Across all chains, DefiLlama counts 6.667 billion and 4.703 billion.
Liquity: the CDP with no governance over rates
Liquity takes the opposite path to Sky. Its contracts cannot be upgraded, and governance does not set any borrowing parameter. It has two versions running side by side.
Liquity v1 and LUSD: no interest, hard redemptions
Liquity v1 launched LUSD in April 2021 (DefiLlama’s first data point is 5 April 2021). You lock ETH, the only collateral, keep a ratio of at least 110%, and pay no interest, only a one-off borrowing fee between 0.5% and 5%, according to the Liquity documentation. Liquidations are absorbed by the Stability Pool: LUSD holders who deposit there see their LUSD used to repay the bad loan and receive the ETH collateral in exchange, usually at a profit.
The dollar door is the redemption: anyone can hand LUSD to the protocol and receive one dollar’s worth of ETH, minus a fee of at least 0.5%, taken from the riskiest loans. That gives LUSD a hard floor. On 28 October 2026 its supply had shrunk to about 26 million and Liquity v1 held 168 million dollars of collateral, per DefiLlama.
Liquity v2 and BOLD: borrowers choose their own rate
BOLD, the v2 coin, keeps the redemption floor and adds interest, but nobody votes on it. Each borrower picks a rate, from a 0.5% minimum upward, and redemptions hit the loans paying the lowest rate first. The rate therefore works as protection: pay more than your neighbours and you are less likely to be redeemed. When BOLD trades above a dollar borrowers can lower their rates; when it trades below, redemptions become profitable and borrowers raise their rates to stay out of the queue.
The collateral is ETH, wstETH and rETH, each in its own branch with its own Stability Pool. Positions are called Troves, are transferable as NFTs, and need a minimum debt of 2,000 BOLD. The minimum collateral ratio is 110% on ETH and 120% on the two staking means depositing a proof of stake coin with a protocol that stakes it for you and issues a transferable token (such as Lido stETH or rETH) representing the deposit plus rewards, which you can use or sell while the coins stay staked.">liquid staking tokens. Opening a loan costs an upfront fee of seven days of the branch’s average interest, and changing your rate within seven days of the last change costs the same, so that borrowers cannot dodge redemptions by jumping rates at the last minute.
Of the interest paid, 75% goes to Stability Pool depositors and 25% to liquidity incentives chosen by LQTY holders. Liquidation pays the pool a 5% penalty on top of the debt. Instead of v1’s Recovery Mode, v2 has two safety switches. Safety Mode restricts some actions on a branch while its total collateral ratio is below 150% (ETH) or 160% (wstETH, rETH), and lifts them once the ratio recovers. Shutdown is the permanent one: if a branch’s total collateral ratio falls below 110% (ETH) or 120% (wstETH, rETH), or its price oracle fails, that branch stops new borrowing for good and is wound down through special redemptions.
Here is the redemption example from the Liquity v2 documentation, with ETH at 3,000 dollars. A Trove holds 10 ETH and 20,000 BOLD of debt. A redeemer hands in 15,000 BOLD. The Trove’s debt drops to 5,000 BOLD and the redeemer takes 5 ETH (15,000 dollars) minus the 0.5% fee, which stays in the Trove. After the redemption the Trove holds 5.025 ETH and 5,000 BOLD. The borrower lost nothing in dollar terms, and even gained 0.025 ETH, but lost half of their ether exposure without asking.
The fee is min(0.5% + baseRate, 100%). The baseRate rises with each redemption in proportion to its size relative to supply and decays with a six-hour half-life, so a wave of redemptions becomes more expensive while it lasts.
The launch was rough. On 12 February 2025, weeks after the first v2 deployment, the team reported a possible Stability Pool issue. It first asked Stability Pool depositors, then borrowers and BOLD holders, to close their positions, and said on 13 February that to its knowledge no users had been impacted (Liquity blog). A redeployed v2 went live on 19 May 2025. On 28 October 2026 BOLD supply stood at about 38 million and Liquity v2 held 94.3 million dollars of collateral, per DefiLlama. Since DefiLlama’s series for the new BOLD starts on 27 May 2025, its daily price has stayed between 0.9925 (8 June 2025) and 1.0063 (9 March 2026).
Protocol stablecoins: Aave’s GHO and Curve’s crvUSD
The last two designs come from protocols that already had something else to sell: a lending market for Aave, a stablecoin exchange for Curve. Their coins use that base.
GHO: minted by borrowing on Aave
GHO is not minted from an isolated vault. You supply collateral to the Aave v3 Ethereum market, as for any loan on Aave, and borrow GHO, which is created on the spot rather than taken from lenders. Aave’s GHO documentation calls each minting source a “facilitator” with a mint cap set by governance: the Aave market itself, a flash minter, and the GHO Stability Module, which swaps GHO against approved stablecoins at a fixed 1:1 price, a PSM by another name.
The borrow rate is set by Aave governance rather than by the market’s utilisation curve. This figure is our own reading, not a rate published by Aave: the Aave v3 pool contract on Ethereum returned a variable GHO borrow rate of 4.50% a year (APR) on 28 October 2026. Holders can deposit GHO into sGHO, an ERC-4626 vault with instant deposits and withdrawals that pays a fixed rate written in the contract. Its target rate read 450 that day, which we take as 4.50% a year in basis points, with roughly 246.7 million GHO deposited against a cap of 400 million. Check the Aave app before relying on either number. GHO supply was about 698 million (DefiLlama). Its daily price fell as low as 0.9581 on 17 November 2023, a reminder that a young coin can trade under its peg for a while.
crvUSD: soft liquidations through LLAMMA
crvUSD, live since May 2023, is a classic CDP with one unusual part: the liquidation engine. Curve calls it LLAMMA, the Lending-Liquidating AMM Algorithm. When you open a loan, your collateral is spread across 4 to 50 price bands of your choice, much like a concentrated liquidity position on Uniswap v3, according to the Curve developer documentation.
If the collateral price enters your bands, arbitrage traders swap your ether for crvUSD step by step: a soft liquidation. If the price recovers, they swap it back: a de-liquidation. Your loan is never closed in one go at a single price. A hard liquidation, where someone repays the debt and takes the collateral, only becomes possible when the loan’s health falls below zero. The catch is that each back-and-forth costs a little; Curve’s docs say losses are lower with more bands, more liquidity and efficient arbitrage, and a swap fee paid by arbitrageurs partly offsets them. Interest keeps eating health at the same time.
The borrow rate follows a formula: a base rate multiplied by an exponential term that rises when crvUSD trades below a dollar and falls when the PegKeepers hold a large share of debt. The documentation gives a useful reading of its sigma parameter: it is the time over which a borrower pays for the current depeg through higher interest. Governance has retuned the policy several times. According to Curve’s March 2026 recap, vote 1364 brought average crvUSD borrow rates from about 11.1% at the start of that month to about 5.6% at its end, and vote 1358 added a PegKeeper paired with Aave’s GHO, with a 3 million crvUSD debt ceiling: two CDP stablecoins now prop each other up.
PegKeepers are the crvUSD version of a dollar door. They hold pre-minted crvUSD and are allowed only two actions: deposit it into a pool paired with a fiat-backed stablecoin when crvUSD trades above a dollar, and withdraw it when crvUSD trades below. Anyone can trigger them and earns a share of the profit. The docs flag the dependency openly: trouble at USDC or USDT, which feed crvUSD’s dollar price, could push the PegKeepers to dump all their crvUSD into the pools. crvUSD supply was about 222 million on 28 October 2026 (DefiLlama), and its daily price dipped to 0.9849 on 7 March 2024 and to 0.9864 on 6 February 2026.
Side by side
| System | Collateral (examples) | Minimum ratio | Price of borrowing | Liquidation | Dollar door | Savings rate (28 Oct 2026) |
|---|---|---|---|---|---|---|
| Sky (DAI, USDS) | ETH, wstETH, SKY, others | 120% to 175% | Stability fee by vote, 7.41% to 15.5% | Dutch auction, 13% penalty | LitePSM with USDC, no fee | DSR 1.25%, SSR 3.80% |
| Liquity v1 (LUSD) | ETH | 110% | No interest, one-off fee 0.5% to 5% | Stability Pool | Redemption against ETH | None (Stability Pool gains) |
| Liquity v2 (BOLD) | ETH, wstETH, rETH | 110% / 120% | Set by each borrower, from 0.5% | Stability Pool, 5% penalty | Redemption, lowest rate first | 75% of interest to Stability Pool |
| Aave (GHO) | Assets of the Aave v3 Ethereum market | Per asset in Aave | Set by governance, 4.50% APR (our reading) | Aave liquidation with bonus | GHO Stability Module | sGHO, target 4.50% (our reading) |
| Curve (crvUSD) | ETH, wstETH, others | Per market | Formula tied to peg and PegKeeper debt | LLAMMA soft, then hard below zero health | PegKeepers | Not covered here |
Sources: Sky and Aave contracts read on-chain, Liquity and Curve documentation, DefiLlama; all on 28 October 2026. Rates change by vote or by formula: check before relying on any of them.
Where collateralised stablecoins break
Over-collateralisation protects the coin, not necessarily the borrower, and not against every failure. The weak points below come from the designs themselves and from what has already gone wrong.
- When prices fall faster than liquidations can sell, the protocol is left with debt larger than its collateral, known as bad debt. Black Thursday is the textbook case.
- Every system above needs a price feed, an oracle, to know when to liquidate. A wrong or stale price can liquidate healthy loans or leave bad ones open; a later lesson covers oracle design and failures.
- The USDC in Sky’s LitePSM, the stablecoins behind the GHO Stability Module and the pools used by crvUSD’s PegKeepers all tie these coins to Circle or Tether, including their power to freeze tokens.
- Sky’s stability fees and savings rates, Aave’s GHO rate and Curve’s monetary policy parameters are set by token-holder votes. A vote can make borrowing more expensive overnight. Liquity v2 avoids this but cannot fix a bad parameter either.
- A contract flaw can surface after deployment and after audits, as Liquity’s own relaunch shows. Upgradeable tokens such as USDS add a different trust question: who can change the code, and how fast.
- On Liquity, a borrower paying a low rate can see part of their collateral swapped for debt repayment at any time.
The price data shows what stress looks like in practice. These are DefiLlama’s daily price points, not intraday lows, which were sometimes deeper:
| Coin | Date | Daily price (dollars) | Context |
|---|---|---|---|
| DAI | 13 March 2020 | 1.0585 | Above peg after Black Thursday |
| DAI | 12 March 2023 | 0.9669 | USDC depeg |
| GHO | 17 November 2023 | 0.9581 | Early months of the coin |
| crvUSD | 7 March 2024 | 0.9849 | Lowest point in the series |
| BOLD | 8 June 2025 | 0.9925 | Lowest point since relaunch |
| crvUSD | 6 February 2026 | 0.9864 | Before the March 2026 rate cut |
Rules for US and European readers
In the United States, the GENIUS Act (Public Law 119-27, signed into law on 18 July 2025) sets up a regime for payment stablecoins issued by permitted issuers. Its section 14 orders the Treasury, working with the Federal Reserve, OCC, FDIC, SEC and CFTC, to study non-payment stablecoins, including “endogenously collateralized” ones that rely solely on another asset of the same issuer, and to report to Congress within 365 days of enactment, that is by July 2026 (text of the Act on govinfo). DAI, USDS and BOLD are backed mainly by assets their protocols did not create, such as ETH and USDC, so that label does not fit them neatly, but we found no official classification of any of them under the Act, nor a published report from that study as of 28 October 2026.
On taxes, the IRS has not, to our knowledge, issued guidance specific to opening a vault, minting DAI or converting DAI to USDS. General US crypto rules are in our guide to crypto taxes in the United States; for vault positions, ask a tax professional before assuming a borrow or a swap is tax-free.
In the European Union, MiCA regulates stablecoin issuers and crypto service providers, but DAI and USDS have no single company issuing them. Whether they count as e-money tokens is disputed: Ledger Insights reported on 28 January 2025 that a legal report commissioned by Sky token holders concluded the e-money token rules do not apply to DAI or USDS, so EU exchanges could list them. An adviser to the European Commission who worked on MiCA took the opposite line in the same article and said that, in case of doubt, European courts would have to decide. Our MiCA explainer sets out the general rules.
Check the numbers yourself
Every figure in this lesson can be reproduced without an account or an API key. Three habits cover most of it.
- On Etherscan, open the Pot contract (0x197E90f9FAD81970bA7976f33CbD77088E5D7cf7) and read dsr, a per-second rate with 27 decimals; raise it to the power 31,536,000 and subtract one to get the annual DSR. The sUSDS contract does the same for the SSR with ssr.
- On the same site, check the USDC balance of the LitePSM pocket quoted at the top of this page and compare it with DAI plus USDS supply on DefiLlama’s stablecoins page.
- To practise without risking money, use the Sepolia test network with test ETH from a faucet: Aave’s GHO documentation lists Sepolia contract addresses, and Liquity’s documentation does the same for v2. Test tokens have no value, and gas there costs nothing real.
The next lesson moves to a different kind of crypto dollar: USDe, which holds its value with a hedged derivatives position instead of a loan, and fails in different ways.
