The third-largest stablecoin almost never leaves its own system

Crystal Foresight tracks stablecoin supply, transfer activity, and holder composition across the market. This report covers USDS and its predecessor DAI, together representing $11.1B in circulating supply backed by the same underlying system.

Key takeaways

• The $11.1B headline combines two tokens. USDS ($7.1B) and its predecessor DAI ($4.1B) are both backed by Sky’s Vat system. Of that combined supply, 57.5% – $7.4B – is minted through Allocator credit lines: governance-approved facilities to Sky’s own sub-protocols, backed from lending operations rather than locked reserves.

• 84% of circulating USDS never leaves Sky’s own contracts. Sky’s savings vault (sUSDS) holds $5.9B. Two further staking contracts account for most of the rest. Sky’s own infrastructure holds virtually all of the supply it issues.

• $350.9B in 90-day transfer volume is mostly internal recycling. More than half – $189.8B, or 54% – went to four Sky-controlled destinations. USDS turns over 31 times its supply in 90 days. That multiplier is nearly entirely internal.

• Retail savers hold less than 1% of the savings vault. Crystal mapped current sUSDS balances across all holders. Sky’s own protocols hold 32%. Four unlabeled addresses each holding over $200M account for a further 34%. Retail wallets – the apparent target audience for the savings product – hold $32M across 2,824 wallets: less than 1% of total value locked (TVL, meaning the total assets deposited in the vault).

How USDS is backed

Most major stablecoins follow a simple template: an issuer holds reserves and issues tokens against them. Every dollar of USDC is backed by a dollar held in reserve. The reserves exist first; the stablecoin follows.

USDS departs from that template for more than half its supply.

The reserve-backed portion

The reserve-backed portion works as described. Sky’s Peg Stability Module (PSM) holds $4.8B in USDC deployed into short-term government securities: users swap USDC for USDS at 1:1, and the USDC stays in reserve. Overcollateralized vaults – where users lock ETH, wstETH, and WBTC to borrow USDS – contribute another $549M. Real-world asset vaults, backed by tokenized US Treasuries and credit facilities, account for $106M. These positions are enforced in Sky’s core accounting contract, the Vat, where collateral is locked and liquidated automatically if it falls below threshold.

The Allocator model

The other 57.5% – $7.4B – works differently. These are Allocator positions: Sky governance votes to extend a debt ceiling to a sub-protocol, which then mints USDS against that ceiling. Spark received a $4.1B ceiling; Bloom, $2.7B; OBEX, $608M. At the Vat level, no hard collateral backs these positions. The Vat records the obligation against a governance-issued ceiling, not a locked asset.

One level down, real backing does exist. Spark lends its minted USDS through SparkLend, where borrowers post overcollateralized ETH and wstETH. Bloom and OBEX deploy into T-bills and structured credit. But consider what that means for Spark: a borrower posts ETH to borrow USDS from SparkLend – that ETH secures the borrower’s loan. Spark itself, however, minted that USDS against a governance credit line, implicitly backed by the same loan book. The ETH is doing double duty: it is the borrower’s collateral and the de facto reserve behind the USDS in circulation.

The difference from a standard vault is not whether collateral exists. It is how that collateral is enforced. A standard vault’s collateral is locked in the Vat and liquidated automatically if it falls below threshold – trustless and instantaneous. Allocator backing sits outside the Vat: losses flow back to Sky with no automatic on-chain mechanism to intercept them. T-bills and structured credit cannot be liquidated on-chain in seconds the way ETH can, adding liquidity risk if redemptions spike.

A dominant actor in SparkLend

The on-chain data puts a specific face on this structure. SparkLend’s 792 borrowers have posted $5.2B in collateral against $586M in outstanding debt – a system-wide ratio of 8.9 times, far above the 1.5 to 2 times typical of leveraged decentralized finance (DeFi) borrowing. DeFi refers to financial protocols that run on blockchains without a central intermediary.

One position dominates. A single address holds $3B in collateral while borrowing just $71.6M in USDS – a 42 times ratio. A second address, confirmed on Etherscan as Abraxas Capital Management – a London-based asset manager that describes itself as managing $3B, whose core strategy is ETH-collateralized stablecoin arbitrage – holds a further $547M. Together the two addresses account for over two-thirds of all SparkLend collateral.

The profile – large collateral, minimal borrowing – suggests a firm using SparkLend as a collateral management facility: ETH parked, a small USDS position held for optionality, capital deployed elsewhere. These are not retail users.

At a 42 times collateral ratio, SparkLend’s liquidation mechanism is effectively irrelevant: ETH would need to fall 97% before the position approached the liquidation threshold. Well before that point, a holder of this size can simply withdraw. Trustless liquidation protects against involuntary default. It does not protect against a dominant holder choosing to exit.

There is a further implication. Spark’s $4.1B Allocator ceiling is sized against its collateral base, but SparkLend has only deployed $586M of that as actual loans – a 14% utilization rate. The $3.5B gap between the ceiling and the outstanding loan book presumably represents minting capacity available to Spark beyond what its lending operations require, backed by the same borrower ETH, with no on-chain constraint on how it is deployed.

Whether the Allocator model represents a considered architectural choice or an underappreciated structural risk depends entirely on what happens when redemptions spike and governance moves slowly. The track record through the 2022 crash, the 2023 USDC depeg, and multiple governance crises is real. So is the absence of a trustless mechanism to intercept losses.

How USDS is used

Once issued, USDS does not circulate broadly. Almost all of it stays inside Sky’s own contracts.

84% of circulating USDS – $5.9B – sits in Sky’s savings vault (sUSDS). A rewards contract distributing USDS to SPK rewards holds another $831M (12%). A third Sky staking contract holds $207M. Combined, Sky’s own contracts account for virtually all circulating USDS.

The 90-day transfer record reinforces this. $350.9B flowed in gross transfer volumes – the kind of figure cited in reports about trillions flowing through stablecoins – which at face value implies broad circulation. But more than half went to Sky-controlled destinations: Spark received $86.5B (26%), Sky Savings $43.2B (13%), MakerDAO governance $36B (11%), and Sky’s treasury $24.1B (7%). Together, those four destinations received $189.8B – 54% of all 90-day transfer volume. USDS turns over 31 times its supply in 90 days. That multiplier is nearly entirely internal.

The contrast with other major stablecoins is direct. USDT moves through tens of thousands of wallets, exchanges, and payment processors globally – its volume reflects genuine breadth across diverse entities and use cases. DAI, Sky’s predecessor stablecoin, circulates across a far wider set of external DeFi protocols and individual wallets. RLUSD shows transfer patterns consistent with payment network activity distributed across a broad and varied recipient base. The $350.9B headline for USDS reflects something different: a single protocol’s internal machinery running.

What sUSDS actually serves

Sky’s pitch for sUSDS is simple: deposit USDS, earn the Sky Savings Rate (currently 3.65% APY), withdraw any time. Depositors who held from launch have earned 9.77% in 20 months – verifiable directly from the contract. That is a real and functional yield product.

The question is who actually uses it.

Crystal mapped current sUSDS balances across all holders using on-chain transfer data:

sUSDS holder composition by segment. Source: Crystal Intelligence on-chain analytics. 

Sky’s own protocols account for the largest identifiable slice. Spark holds $1.4B in current sUSDS balance – against $10.7B in lifetime gross deposits across 9,488 transactions, cycling capital between lending operations and the vault. Sky’s own treasury shows $7.0B in gross deposits against just $109M in net deposits. That gap between gross flows and current balances drives the $41.8B in 90-day sUSDS inflows. This is active capital management, not new money entering the system.

Four addresses each hold over $200M in sUSDS – together $1.7B, a third of the entire vault. On-chain data cannot confirm who these holders are, but positions at that scale point to institutional treasury management.

Everything below the large-holder tier – labeled protocols, the 654 mid-sized whales likely representing institutional activity, and retail – adds up to roughly 34% of TVL. That is the share consistent with genuine end-user demand outside Sky’s own protocols.

Retail – individual users using sUSDS as a savings account, the product’s apparent target audience – holds $32M across 2,824 wallets: less than 1% of TVL. In practice, the savings vault is a treasury management tool.

What the blockchain reveals

The $11.1B figure reported for USDS combines two tokens backed by the same system. More than half of that supply is minted through governance-issued credit lines, not locked reserves. Virtually all circulating supply sits in Sky’s own contracts. And the savings vault marketed to retail depositors is held almost entirely by Sky’s own protocols and a small number of large institutional addresses.

These are characteristics, not necessarily flaws. But they matter for anyone assessing the instrument: a compliance team evaluating counterparty exposure, an institution considering USDS as collateral, or a retail user comparing savings options.

Three questions are worth tracking: Does Spark’s 14% utilization rate on its $4.1B Allocator ceiling increase, and what backs the expansion if it does? Do the four unlabeled mega-whale positions in sUSDS show signs of coordinated movement? And does retail participation grow beyond 1% as the savings rate adjusts?

Data as of May 25-26, 2026. All figures derived from Crystal Intelligence on-chain analytics and Dune Analytics. USDS collateral breakdown from the MakerDAO/Sky Vat contract. Market caps and transfer volumes are approximate. This report is for informational purposes only and does not constitute investment advice.