Introduction

Dollar-pegged stablecoins have grown to $300 billion outstanding as of August 2026, of which reserve-backed stablecoins make up roughly $270 billion. That supply is largely concentrated in two coins, but it hides a great deal of variation. Tether’s USDT and Circle’s USDC account for 95 percent of the reserve-backed total, yet their footprint is spread across many blockchains and venues (e.g., centralized exchanges, lending protocols, decentralized exchange pools, bridge contracts, and self-custodied wallets).

This note maps reserve-backed stablecoin supply across blockchains and venues, using granular protocol-level data from 2023 through July 2026. Mapping stablecoin supply helps us understand holder behavior, current use cases, and potential risks. For example, it can inform how much of the stablecoin supply supports crypto trading or other uses. It could shed light on the extent of stablecoin-funded leveraged positions through balances supplied to lending protocols and collateralized debt positions (CDPs), and therefore their contribution to fragility in the crypto ecosystem. It could also help gauge stablecoin run risk if, for example, stablecoin balances on some venues are more volatile than others. It also speaks to whether stablecoins primarily function as dollar balances held for reasons other than yield, or as assets that generate returns.

The venue mapping in particular addresses the last question directly because venues differ in whether they offer remuneration. Under the GENIUS Act, stablecoin issuers cannot pay interest or yield directly to holders. Exchanges and protocols face no such rule to date, although the CLARITY Act, still under discussion, could change that. If holders were strongly yield-seeking, stablecoin balances in yield-bearing venues should be higher. However, roughly half the supply is held in wallets we cannot match to any venue. We cannot observe whether these balances earn yield, but since issuers generally do not pay interest, the most plausible reading is that many holders do not pursue the yields available to them.[1]

We document seven key facts:

1. The intermediated share of stablecoin supply is about one-half, and rising. Of $269.4 billion outstanding on 28 July 2026, $128.9 billion, or 47.8 percent, sit at centralized exchanges and inside decentralized finance (DeFi) protocol contracts (the intermediated share). The remaining 52.2 percent reside at addresses we cannot attribute to any venue (the unlocated share). The intermediated share rose from 42.8 percent on average in 2023, suggesting a growing proportion of stablecoins are put to work within the crypto ecosystem rather than migrating to uses outside it, such as real-world payments.
2. Most stablecoins likely earn no return, suggesting holders are not primarily yield-driven. Only 9 percent of stablecoin balances are supplied to venues that explicitly pay holders, namely lending protocols and decentralized exchange (DEX) pools. A further 52.2 percent cannot be matched to any venue. We cannot verify that all these unlocated balances go unremunerated since some may sit at exchanges that do not disclose their addresses, though issuers themselves do not pay interest.
3. The majority of located stablecoins are held on exchanges. Centralized exchanges hold $76 billion, compared to $53 billion across all DeFi and bridge categories combined. And exchange balances could be understated if some unlocated balances belong to exchanges that do not publicly disclose their addresses (though holder-level evidence suggests this is unlikely to account for a large share). Lending protocols hold $21 billion, bridge contracts $17 billion, $8 billion in other DeFi protocols, and $3 billion in DEX pools. These categories are defined in Section 2.
4. USDT dominates exchanges while USDC dominates DeFi. USDT accounts for 81 percent of all stablecoins held at exchanges. USDC accounts for 62 percent of all stablecoins supplied to DeFi protocols, excluding bridges.
5. The distribution of stablecoins differs sharply by blockchain. About 70 percent of the USDT on Ethereum sits on exchanges. As for USDT on Tron, only 4.6 percent sits on exchanges with 93 percent residing in unlocated wallets. These are the same stablecoin, the same issuer, governed by the same redemption right. The only differences is the blockchain it circulates on. USDC varies across chains too, but less dramatically and is more concentrated, with 64 percent of supply circulating on Ethereum.
6. Flat aggregate stablecoin supply can mask large reallocations across venues. Since October 2025, total stablecoin has remained essentially flat. Underneath, however, roughly $20 billion left DeFi protocols and moved to exchanges and unlocated wallets. Aggregate supply, the measure most often cited as a gauge of stablecoin adoption, cannot capture allocative shifts of this kind.
7. Venue composition shapes how stablecoin runs propagate. In March 2023 when USDC de-pegged, exchange balances made up 15.2 percent of USDC supply but accounted for 40 percent of the total outflow. Redemption pressure concentrates at exchanges, whether because holders are quicker to exit or because exchanges are the main off-ramp for those who cannot redeem directly with the issuer. Either way, the venue composition is informative of stablecoin outflow risk under stress.

We map the stock of stablecoin supply to where it resides, in contrast to a number of studies mapping stablecoin activity based on flows or transfer volume (Batra et al., 2026; Cong et al., 2026; Higginson et al., 2026; Schär et al., 2026). Two papers are closest to ours. Noll (2026) uses aggregate reports from the same DefiLlama source to estimate the distribution of stablecoin supply across use cases at a single date, and then allocates the remainder across payments, transfers, and idle balances using external survey estimates and assumptions about the velocity of money. Our measured shares replicate his November 2025 shares to within roughly two percentage points across each venue category, but we do not infer which balances are used for payments and instead report the unattributed share as a residual. Gopinath (2026) uses commercial data to split holdings between centralized exchanges and a combined self-custody-and-other category and additionally decomposes transfer flows on Ethereum. Because her non-exchange category includes DeFi protocol balances, our finer venue breakdown shows how much of it sits in identifiable smart contracts rather than in externally owned wallets. This is particularly useful for assessing leverage, run risk, and yield-seeking behavior within the stablecoin ecosystem. Relative to both, we build estimates from granular historical data across twelve reserve-backed coins, six blockchains, and coin-chain pairs, while relying solely on public data sources.

Data and preliminaries

A stablecoin is a token that transacts on a blockchain, and the blockchain is a public ledger that tracks which address holds how much. A stablecoin issuer, such as Tether or Circle, can mint the same stablecoin on several blockchains. USDT circulating on Ethereum and USDT circulating on Tron are the same claim on the same issuer, but they are separate balances on separate ledgers. Moving between them requires the issuer to destroy tokens on one chain and create them on another. Every stablecoin unit in circulation therefore sits at some address on some chain. Those addresses fall into three broad groups.

The first is centralized exchanges. These are traditional, ‘off-chain’ companies (e.g., Binance, Coinbase, Kraken) that hold customers’ stablecoins on their behalf at addresses controlled by the exchange. Because these firms typically know their customers, balances held there are identifiable in a way that other on-chain balances are not.

The second is DeFi protocol contracts. These are programs that run directly on the blockchain and hold tokens deposited by users. Unlike exchanges, they do not take custody through a firm that conducts customer verification. A lending protocol accepts deposits and lends them out, paying the depositor interest. A DEX liquidity pool holds two assets and lets anyone trade between them, paying providers a share of trading fees. A CDP lets someone lock stablecoins or other cryptocurrencies to borrow a different token (often a stablecoin). A bridge contract locks tokens on one chain so a matching, “wrapped” token can circulate on a different chain, and the locked token backs the wrapped token. CDPs and bridges do not pay interest on the supplied stablecoins, but the stablecoin minted on top of the CDP or the wrapped stablecoin created by the bridge could be subsequently supplied to a yield-bearing protocol.

The third group is composed of addresses that belong to neither (i.e., a residual). These may be individuals holding their own tokens, firms holding their own balances, or intermediaries that have not disclosed their addresses. We call these balances “not located,” because we can see the tokens but their respective addresses are not matched to any venue. This note measures how the outstanding supply of stablecoins divides across these three groups, separately for each coin and each blockchain. Figure 2 illustrates this mapping.

Data coverage

We use daily data from DefiLlama covering 1 January 2023 to 28 July 2026. The data includes market capitalization by coin and blockchain, which we use to measure supply. We also collect data at the protocol level that identifies the value of stablecoins held in identified venue contracts and addresses, by coin, blockchain, and venue type, as well as balances sitting on centralized exchanges. We consider twelve reserve-backed dollar stablecoins: USDT, USDC, USD1, USDG, PYUSD, RLUSD, FDUSD, TUSD, BUSD, GUSD, USDP and USAT. Noll (2026) also draws on DefiLlama, using the platform’s aggregated protocol report rather than the underlying data, and reports a single-date snapshot.

We specifically exclude crypto-backed and synthetic stablecoins such as DAI, USDS and USDe.[2] Stablecoins pegged to other currencies or to assets such as gold are also excluded. However, when a reserve-backed stablecoin is posted as collateral inside a protocol that issues one of these tokens, the balance is captured in the venue data (e.g., CDPs).

We focus our coverage on six blockchain buckets: Ethereum, Tron, Solana, BNB Chain, Polygon, and an aggregate “Other” bucket that includes Base, Arbitrum, Avalanche, TON, among others.

Venue categories

We sort located protocol balances into six categories:

• Centralized exchanges (CEX): balances at publicly disclosed addresses belonging to centralized exchanges. These are trading venues that hold customer assets in custody. Someone might hold stablecoins on a CEX to maintain balances to trade against cryptocurrencies without needing to move funds on-chain frequently. Some exchanges offer rewards for holding balances of specific stablecoins with them, but many do not.
• Lending protocols: tokens supplied to on-chain lending markets (e.g., Aave or Compound). A depositor supplies stablecoins to a pool, and borrowers post other crypto as collateral to draw from it. The depositor earns interest set by the utilization of the pool (i.e., coins borrowed as a percentage of total coins supplied). This is the most direct way to earn yield on stablecoins on-chain. We do not count borrowed amounts, so a dollar that is supplied and then borrowed is not double counted.
• Bridge contracts: tokens locked as backing for a wrapped version of the stablecoin on another chain.[3] Because an issuer mints stablecoins directly on some chains, moving a stablecoin elsewhere usually means locking the original coin in a bridge smart contract while a matching token, called a wrapped token, is created on the destination chain. The locked coin is collateral for the wrapped version. The locked coin earns no interest and cannot be transferred while the wrapped token is outstanding, though the wrapped token itself could be put to productive use.
• Collateralized debt positions (CDP) and collateral: reserve-backed stablecoins (or other tokens) are posted as collateral to mint a different token against them, typically a crypto-collateralized stablecoin like DAI or USDS. This includes peg-stability modules, which swap a reserve-backed stablecoin one-for-one with the protocol’s own token. In this case, the stablecoin holder is a borrower rather than a lender, and the posted collateral does not earn a return and may incur fees. A return, if any, comes from what the borrowed token is subsequently used for.
• Decentralized exchange (DEX) liquidity pools: tokens committed to automated market makers such as Uniswap or Curve. At least two assets, one being a stablecoin, are deposited into a pool wherein anyone can trade one of the assets for the other at an algorithmically set price. The depositor of the asset pair earns a share of the trading fees.
• Other DeFi: fifteen remaining protocol types categorized by DeFiLlama, listed in Appendix B. These include yield strategies that route stablecoin deposits into other protocols, collateral posted against derivatives and options positions, synthetic assets, insurance pools, payment protocols, and tokenized real-world assets. Most pay a return, but not all.

We define the difference between market capitalization (i.e., total supply) and the sum across these six categories as “not located.” These are often referred to as externally owned accounts (EOAs). Note that it is not necessarily a measure of self-custody although it does include balances held in self-custodied wallets. But it could also include any stablecoin held at an exchange that has not published its address, along with balances in wallets that are not classified.

One important caveat involves data on centralized exchange balances. Specifically, centralized exchange coverage depends on voluntary disclosure. DefiLlama adopts the standard practice of building exchange balances from publicly available address lists that exchanges typically report themselves. Consequently, stablecoin balances held on an exchange that does not report its addresses would instead be counted under ‘not located’. Even those exchanges that do publish may not publish all their addresses. We therefore interpret exchange balances as a lower-bound estimate.

Data cleaning and robustness

We clean the raw files sourced from the DeFiLlama public API with several steps outlined in Appendix B. Aggregated metrics that we are concerned with are not materially impacted by our data processing approach, as we show that the intermediated share (stablecoins sitting across all located venues) lies between 47.5 percent and 47.8 percent under multiple data-processing approaches, including using the raw files with no cleaning at all.

Mapping aggregate stablecoin supply

Figure 1 and Table 1 report the allocation of reserve-backed stablecoins in aggregate. Of $269.4 billion outstanding, $128.9 billion is intermediated (i.e., supplied to DeFi contracts or held on CEXs).[4]

The other $140.5 billion, or 52.2 percent of stablecoin market cap, is held at addresses with no identifiable venue. Some of that is exchange balances at addresses that are not public, so the true unlocated share is lower. But still, most of it is likely stablecoins sitting in EOAs, which means that these holders are not earning interest, as they may be holding stablecoins as dollar balances without seeking a return on them.[5] The largest located venues are centralized exchanges, and while several exchanges offer yield-like rewards for holding stablecoin balances with them, many stablecoins held on exchanges do not earn yield-like rewards.[6]

That said, the total intermediated share of stablecoins has been increasing. It averaged 42.8 percent in 2023 and 51.5 percent in 2026, peaking above 54 percent in March 2026 before pulling back. This is consistent with a growing share of balances being put to work, whether for crypto trading or in DeFi protocols, even if the overall level still suggests the aggregate holder base is not particularly yield-driven. But at the same time, one would expect a declining intermediated share if stablecoins were increasingly being used for non crypto-related, non-speculative activities such as real-world payments. Section 4 shows that among stablecoin-blockchain pairs, some have seen their intermediated share notably decline, while others have seen it grow.

Of the $128.9 billion located supply, $76 billion resides at centralized exchanges. This amounts to 59 percent of the located total and 28.2 percent of all supply. This is the largest attributed share of stablecoins, and consistent with stablecoins’ primary use case being cryptocurrency trading. The total across all DeFi categories is $52.9 billion, meaning centralized exchanges hold 1.4 times as much as the entire decentralized finance ecosystem. Due to the way centralized exchange balances are reported, the exchange figure is a lower bound, so the stablecoin balance on exchanges may be even larger. Our estimated 28.2 percent share closely aligns with share estimates of 26.4 percent as of November 2025 from Noll (2026) and CryptoQuant’s 27 percent as of July 2026. DEX pools hold about $3.2 billion or 1.2 percent of stablecoin supply, though this category is the least reliably measured in our data (see Appendix B).

Bridges hold $16.8 billion or 6.2 percent of stablecoin supply.[7]This is the least productive of the large categories in the data. A stablecoin locked in a bridge is not lent to anyone, funds no position, and earns nothing. It instead backs a wrapped version circulating on another blockchain (often one that the stablecoin issuer does not service). That said, the wrapped version could then be placed within an economically productive DeFi protocol on the destination chain.

The productive core of stablecoin DeFi includes lending and CDPs. These categories together hold $25.5 billion, or 9.5 percent of stablecoin supply.

All these categories show significant movement over time. Figure 1 shows that exchange balances have roughly doubled since January 2023, from $36.8 billion to $76.0 billion. Lending has grown faster in proportional terms, from $8.9 billion to $21.0 billion. Bridges have grown from $0.7 billion to $16.8 billion, the fastest-growing category in the data and now larger than every DeFi category except lending.

Venue differences across coins and blockchains

The aggregate distribution of stablecoin supply conceals at least two kinds of variation. First, the two major stablecoins, USDT and USDC, are used differently from each other. Second, the same stablecoin is used differently depending on the blockchain it circulates on.

Variation across coins

Figure 3 reports the composition of USDT and USDC supply across venue categories. USDT accounts for 81 percent of all stablecoins held at exchanges, while USDC accounts for 62 percent of all stablecoins supplied to DeFi protocols, excluding bridges. The figures also show that this contrast remains present as a share of each coin’s own supply. About 33.4 percent of USDT is held at exchanges against about 4 percent in lending protocols and CDPs, whereas 13.9 percent of USDC is held at exchanges against 22.4 percent in lending protocols and CDPs (Table 2).

The two coins grew over time to their current compositions in different ways. In the case of USDT, total supply rose from $66 billion to $184 billion between January 2023 and July 2026, and exchange balances rose from $18.6 billion to $61.5 billion. Its unlocated share barely moved, from 53.8 percent to 55.3 percent. As a result, new supply split between exchanges and non-located wallets in roughly the proportions that already existed. The one new venue for USDT is bridges, which grew from nearly zero to $9.7 billion.

By contrast, USDC’s supply grew less, from $44 billion to $72.5 billion, but the unlocated share fell from 72.3 percent to 45.6 percent. In early 2023, USDC mostly resided within the unlocated category. It is now a protocol asset, with lending balances more than quadrupling, from $2.8 billion to $12 billion, and CDP balances rising by 70 percent.

The ten smaller coins make up about $12.8 billion, or 4.7 percent of total stablecoin supply, and they also vary across venues. Some coins live almost entirely at exchanges, e.g., USD1 at 83 percent and FDUSD at 89 percent. Payment and institutional coins barely touch exchanges and lean toward lending: 30 percent of PYUSD supply is locked on lending platforms against 5 percent at exchanges, and 24 percent of RLUSD supply is supplied to lending platforms against no measurable exchange balance.[8]

Variation across blockchains

Figure 4 shows that substantial variation in the composition of stablecoin supply is present across blockchains. Different blockchains are comprised of different networks of users, and house different protocols. Stablecoin holders are therefore likely to migrate to a particular blockchain depending on their needs and the size or relevance of the network.

$128 billion of stablecoins are circulating on Ethereum and we can locate 74.5 percent of them. Tron is the second largest stablecoin chain, with $90 billion in stablecoins circulating (nearly all USDT) but we can locate only 7.4 percent. These two chains carry about 80 percent of all stablecoin supply yet they look nothing alike.

The other chains sit in between, with one exception worth noting. BNB Chain holds 8.8 percent of its stablecoin supply in DEX pools, against 1.2 percent market-wide. It is the only chain where committed automated market maker liquidity is a meaningful share of supply (although as we previously noted, this is the least reliably measured category). Figure 5 plots the composition of supply over time for the three largest coin-chain pairs: USDT on Ethereum, USDT on Tron, and USDC on Ethereum. The first two panels provide a historical look at USDT on Ethereum and Tron, respectively, and reveal a stark and persistent difference in how the same stablecoin is held across these chains.

USDT on Ethereum grew from $28 billion to $76 billion between January 2023 and July 2026, and exchange balances grew six-fold from $8.7 billion to $52.8 billion. Meanwhile, its unlocated share fell from 59 percent to 11 percent. A possible interpretation is that USDT on Ethereum is predominantly used for trading.

By contrast, USDT on Tron grew from $32 billion to $90 billion over the same period, but exchange balances fell, from $9.5 billion to $4.1 billion. Its unlocated share rose from 56 percent to 93 percent, and in dollars from $18.7 billion to $83.1 billion. Our comparable estimate, 95.4 percent of USDT-Tron held outside centralized exchanges, aligns closely with the 95 percent self-custody share reported by Gopinath (2026) using a different data source.[9] USDT-Tron houses the largest single pool of unlocated stablecoin supply across all coin-chain pairs. This may indicate growing use of USDT on Tron outside of crypto-related activities. It also means that most USDT-Tron is held as dollar balances rather than in return-generating positions.

One potential concern is that Tron’s unlocated share reflects poor exchange coverage, and that we are mis-attributing unreported exchange balances to the unlocated category. Such a data limitation could then explain the low exchange share and high unlocated share of USDT-Tron. We test this and provide details in Appendix A. Public labels on the Tron blockchain explorer identify only $1.7 billion of exchange-held USDT among the fifty largest addresses[10], which is less than half the $4.1 billion our data tags. Excluding the issuer’s own treasury address, the fifty largest holders of Tron USDT account for 6.8 percent of the total supply on Tron. If the unlocated residual were highly concentrated, the top of the list would be full of unlabeled multi-billion-dollar addresses (e.g., exchanges), but it is not. These findings are therefore consistent with USDT-Tron having a less concentrated holder base and a small balance held on exchanges.

Stablecoin supply and venue-level changes over time

We now consider three periods of important changes to the stablecoin market to understand how balances moved at the venue level. We first examine medium-term boom and stall periods. The period from January 2024 to September 2025 saw the stablecoin market expand by roughly $135 billion. Meanwhile, October 2025 marked the beginning of a period where stablecoin supply remained practically flat. Lastly we study venue-level changes around the contraction to USDC’s supply during the Silicon Valley Bank (SVB) run in March 2023.

The supply boom (Jan 2024 – Sep 2025)

Between January 2024 and September 2025, stablecoin supply grew by roughly $135 billion. This period covered the successful launch of several cryptocurrency exchange-traded funds (ETFs), a crypto bull market that saw bitcoin prices exceed $100,000, and the passage of the GENIUS Act. The left-panel of Figure 6 shows how that supply growth was distributed across venues over this period. $57.5 billion went to unlocated addresses, $27.5 billion to lending protocols, $24.2 billion to exchanges, and $18.0 billion to bridges. During the expansion, venues grew proportionately and this resulted in limited compositional changes.

The crypto winter stall (Oct 2025 – July 2026)

From October 2025 stablecoin supply stopped growing, coinciding with a new ‘crypto winter’. Cryptocurrency markets crashed and entered a bear market, volatility rose, and levered trading positions and looped DeFi strategies rapidly unwound (Ahmed and Natalucci, 2026). Crypto prices remained low through the end of our sample (July 2026), and stablecoin supply remained practically flat over this period (an increase of about $1.4 billion).[11] However, the right-panel of Figure 6 reveals that the underlying venue composition changed substantially despite little change in headline supply. Over this period, exchange balances rose $10.5 billion and unlocated balances rose $11.1 billion, while lending protocols lost $14.6 billion, DEX pools $3.1 billion, and bridges $2.3 billion. Focusing narrowly on aggregate supply numbers would miss what appears to be a major shift out of DeFi.[12]

USDC outflows during the SVB bank run (March 2023)

In March 2023, USDC rapidly de-pegged after Circle disclosed a $3.3 billion reserve position held as deposits with SVB, a bank that at the time was facing a severe depositor run. Figure 7 traces USDC’s total and venue-level supply changes following the event. Over the next 20-30 days following the SVB event, USDC faced outflows of more than $10 billion. Examination at the venue level revelas several additional interesting dynamics.

Consider the first phase, from March 9 to 13, which covers the day before the run, and the official failure of SVB. USDC de-pegged on March 10. From March 10 to 13, total supply fell by $2.7 billion, but unlocated balances fell sharply by $4.7 billion while CDP balances rose $2.3 billion and exchange balances rose $0.6 billion. Some holders did redeem, but a sizable fraction of coins were also used to arbitrage the USDC peg deviation from par by selling discounted USDC into the MakerDAO peg-stability module[13], reflected in the sharp increase in CDP balances. And some holders moved balances from unlocated wallets onto exchanges.

After March 13, redemption persisted and strengthened. USDC supply fell another $8.1 billion, while exchange balances fell $4.9 billion and CDP balances declined, and unlocated balances barely moved. Exchange balances peaked at $7.16 billion on 13 March and fell to $2.26 billion, a 68 percent decline, suggesting that exchanges became the off-ramp for further USDC redemptions.

The venue-level dynamics suggest at least two implications for stablecoin run risk. First, the presence of a peg-stability module could incentivize some holders who would otherwise redeem to instead arbitrage the peg deviation. Second, stablecoin balances on exchanges may be relatively volatile. Measured from the pre-shock baseline of March 9, exchanges held 15.2 percent of USDC supply but accounted for 40 percent of the total decline in supply. This suggests that the venue where a stablecoin is held could inform stablecoin run risk.

Concluding remarks and some economic implications

We use public protocol-level data to estimate the distribution of aggregate stablecoin supply across coins, blockchains, and venues. Several facts emerge about how stablecoins are used, and about the various risks they might carry.

Our findings suggests that many holders treat stablecoins the way they treat cash (as dollar balances held for reasons other than generating returns). Although, it does not necessarily mean that these balances are transactional, as they could reflect idle balances or store-of-value holdings. Nor does it imply that demand is insensitive to yields on outside options. As with physical currency, holdings may respond to the opportunity cost of forgone interest even when the asset itself pays nothing.

The sizable share of stablecoin balances held on exchanges suggests that crypto trading remains a prominent use case. Exchange balances also sit with identifiable intermediaries that conduct customer due diligence, which makes them more traceable and, if necessary, freezable than balances at unattributed addresses. Gopinath (2026) develops the implications, splitting holdings between exchanges and a combined self-custody-and-other category, and arguing that the dominance of the latter means stablecoin usage is drifting toward the most anonymous form of money available, with consequences for tax enforcement, sanctions, and capital controls.

Our venue decomposition adds a qualification. Alongside the $76 billion at exchanges, $53 billion, or 19.6 percent of supply, sits in DeFi protocols and bridge contracts, which her non-exchange category does not separate from self-custodied balances. While DeFi protocols perform no customer due diligence, balances supplied to them are identifiable, not freely transferable, subject to counterparty and smart-contract risk, and in some cases can be frozen. These are not properties of cash, so grouping them with self-custodied balances overstates how closely the non-exchange share resembles currency. The distinction is most material for USDC, where DeFi and bridge balances are roughly half of all non-exchange holdings.

Exchange balances also behave distinctively under stress in ways that could inform run risk. This reflects the responsive nature of exchange balances, and that exchanges are the main redemption channel for holders who cannot redeem directly with the issuer. Either way, exchange balances appear to trade lower compliance risk for higher run risk, though this rests on the single stress episode that we study.

Appendix A. USDT-Tron top holder validation

We estimate a 92.6 percent unlocated share of USDT on Tron, which is by far the largest unlocated pool of stablecoins across coin-chain pairs. It is possible that exchanges underreport addresses that hold USDT on Tron, therefore leading our estimate to mis-attribute exchange balances to the unlocated category. We test this possibility directly using data from Tron blockchain explorer’s public holder rankings, pulled on August 25, 2026. We pull the top fifty holding addresses. Because exchange addresses tend to be very large, exchanges typically show up as top holders.

Public labels identify $1.7 billion of exchange-held USDT in the top fifty Tron addresses (Table A1). Our data tags $4.1 billion on exchanges, which is about 2.4 times as much. This implies that our source captures these exchange balances but also exchange balances held in addresses outside the top fifty.

Excluding the treasury address, the top fifty addresses hold 6.8 percent of USDT-Tron.[14] The largest holder is an exchange hot wallet at $681 million and the fiftieth holds $50 million. The two largest unlabeled addresses hold $500 million each, round to the dollar, suggesting a custodian or single institutional counterparty (rather than aggregated user balances that typically comprise an exchange address).

Appendix B. Data construction and robustness

The underlying data for this analysis was sourced from a granular dataset built using the DeFiLlama public API. This parent dataset was constructed by pulling the /protocol/{slug} endpoint for every protocol listed on DeFiLlama into a panel of 92 million observations. The grain of the parent dataset was on a protocol, token, chain, and date level, with historical total value locked (TVL) and Total Supplied amounts of all crypto-assets dating back to 2017.

Before aggregation, the panel was cleaned to exclude deprecated protocols, malformed chain labels in the source API, and known symbol/ID collisions across stablecoins. The DeFi panel was then built by grouping multiple categories of DeFi activity, stablecoins, and protocols to generate a daily historical panel that tracked how stablecoin supply was being deployed across centralized exchanges, lending, DEX, CDP and other on-chain venues. We considered the following twelve reserve-backed dollar-pegged stablecoins in the analysis: USDT, USDC, USD1, USDG, PYUSD, RLUSD, FDUSD, TUSD, BUSD, GUSD, USDP, USAT.

From the parent panel, two aggregated, daily DefiLlama datasets are produced, one storing market capitalization for each stablecoin-chain pair, and another with venue category values for the same set of stablecoin-chain pairs. Some observations, especially for the younger or smaller coins, appear to suffer from data quality issues (e.g., negative or missing values). We process the data through a simple cleaning procedure to address some of these problems. That said, we find that these data discrepancies are immaterial upon aggregating the data, and that our aggregate estimates are robust to a variety of data processing procedures (Table B1). The following six rules are applied in sequence:

1. Verify no duplicate coin-chain-category-date observations.
2. Verify that blockchain buckets match across both files (six buckets: Ethereum, Tron, Solana, BNB Chain, Polygon, “Other”).
3. Set negative category balances to zero. 100 identified totaling $39 million, almost all in one coin-chain-category group between July and October 2025.
4. Void and interpolate three negative supply observations present for TUSD-Ethereum in May 2022. The largest of these negative values was −$798 million. Retain market capitalization values below $1 million but exclude them from aggregation of the corresponding venue shares. These could reflect coins winding down or pre-launch.
5. Interpolate single-day dropouts. Three identified against a seven-day centered median.
6. Reindex the panel to the full grid after accounting for absent or missing rows. Single-day gaps are filled forward, two-to-seven-day gaps are linearly interpolated, and data gaps longer than seven days are left unfilled and logged.

We also run robustness tests to assess the sensitivity of the aggregate estimates across a number of different data processing procedures, including no processing. This is reported in Table B1. Estimates barely vary, with intermediated share estimates ranging between 47.5 percent and 47.8 percent, whether we directly consider the raw data without cleaning, or we follow the six-step cleaning process above, if we restrict the analysis to just USDT and USDC, if we floor particular balance values, or if we exclude small coins.

Composition of the “Other DeFi” category

Other DeFi aggregates fifteen protocol categories labeled by DeFiLlama: Leveraged Farming, Yield Farming, Liquidity Manager, Derivatives, Options, Options Vault, Synthetics, Prediction Market, Basis Trading, Insurance, Stablecoin Issuer, Stablecoin Wrapper, Payments, RWA, and RWA Lending. As of July 28, 2026, ‘Other DeFi’ totals $7.5 billion, 2.8 percent of supply, of which $5.5 billion is USDC and $4.5 billion lives on Ethereum. It has grown from $1.9 billion in January 2023. We consider this category a catch-all, even though some categories such as RWA are not technically considered DeFi.

DEX liquidity pools

DeFiLlama reports the TVL of DEX/DEX Aggregator pools at the pool level, combining the value of pairs/groups of tokens assigned to the pool. This can create challenges in measuring the true TVL of a stablecoin inside a pool, as the pool-level TVL combines the value of all pooled tokens (.e.g., another stablecoin or crypto-asset being traded within the same pool). It is therefore possible that some of our aggregated observations reflect the combined TVL of all coins assigned to a pool where one leg is a stablecoin. However, our underlying panel is built from DeFiLlama’s /protocol/{slug} endpoint, which provides TVL amounts at the protocol-token-chain level, thereby providing a coin-level value supplied to a single DEX protocol aggregating across its pools. This is distinct from DeFiLlama’s pool-level endpoint (/pools), which reports only a single blended TVL figure per pool and does not support token-level granularity at the pool level. We attempted to verify the protocol-level token attribution empirically. On a given date, the individual per-token USD values for a protocol on a chain sum to the protocol’s reported total TVL for that chain, which suggests that DeFiLlama’s methodology attributes each token its own share of pool value rather than including the full pool TVL across every constituent token paired with the stablecoin. However we cannot confirm whether this methodology was different in the past.

---