How much tokenized value actually works in DeFi

Three counters, one denominator, three different answers

The denominator comes first, because it decides everything after it. Distributed assets are tokens that can leave the issuing platform and move between wallets, as against represented assets, which cannot; chapter 1 sets out that split. Only a distributed token can be deposited into somebody else's smart contract, so it is the only base on which a share of DeFi use means anything. RWA.xyz put the distributed market outside stablecoins at 33.5 bn USD in July 2026, relayed through Stobox's mid-year report, against 369 bn USD of underlying assets that those tokens reference.

The numerator then splits three ways, because a token can be in DeFi in three different senses. Collateral is a token deposited in a lending protocol against which somebody has actually borrowed, as when a fund posts USCC on Aave Horizon and draws USDC against it. A reserve asset is a token sitting on a protocol's own balance sheet behind that protocol's own product, which is what Sky does with BUIDL. A pool asset is a token parked in an automated market maker and available to swap, which supplies secondary liquidity and no borrowing capacity. The first sense is the strongest one, because there the same asset performs a second, independent economic job.

Counting only the first gives 0.73 to 1.01 bn USD, which is 2.2 to 3.0% of the distributed market. That range is the author's own estimate rather than a measurement, assembled from five components of unequal quality, and no data provider publishes a metric for the value of tokenized assets pledged in lending protocols. One of the five components is an admission of ignorance rather than a number: deSPXA on Euler, OUSG on Fluid and JAAA on 3F Protocol are all in use with no published amount, and the estimate carries them as a 0 to 150 mn USD range that on its own drives most of the uncertainty in the total.

Adding the second sense roughly quadruples the result, and adding the third roughly doubles it again.

The broadest counter comes from BeInCrypto citing RWA.xyz for the second quarter of 2026, at 7.4 bn USD and up 200% year on year. It almost certainly holds two things the collateral counter cannot: tokens that are themselves DeFi products, such as syrupUSDC, a share in Maple's loan book worth 1.044 bn USD across 8,257 holders on RWA.xyz on 19 August 2026, and wrappers counted more than once, since ACRED wrapped as sACRED and deposited into a Morpho vault can appear three times. Nobody has published a breakdown of that figure by protocol.

Two checks point the same way. A separate reading of the market by position rather than by value found that 56% of tokenized assets had no activity in a typical week of the second quarter of 2026, on a wider universe of about 60 bn USD, while the same source's own breakdown gives a different answer: of 1,289 assets worth more than 100,000 USD, only 379 recorded any transfer and 910 stood still, which is 70.6%. Those two cannot describe the same set, so they stay here as a 56 to 71% range rather than an average. Part of that stillness is specification rather than failure, a point made by insights4vc on 12 March 2026: a fund with a 5 mn USD minimum and a full whitelist is built to be subscribed, held and redeemed. That defence does not reach the collateral counter, which takes only distributed assets, and an asset that passes the transferability test and is still pledged nowhere is short of demand rather than short of permission.

The loop earns only when the borrow cost sits below the yield

Everything in this chapter runs on overcollateralization, the rule that a borrower pledges more than the loan is worth, and chapter 3 records what it cost to learn: unsecured on-chain credit failed in every case in that material, Maple taking about 54 mn USD of bad debt in 2022 and Goldfinch voting to wind down in June 2026 with roughly 56 mn USD unpaid.

Looping is the strategy built on top of that rule, in four steps: post collateral, borrow a stablecoin, buy more of the collateral, post it again. It runs in production. The Securitize and Gauntlet strategy on Apollo's tokenized credit fund does exactly this, with a vault on Polygon PoS posting ACRED as collateral on Morpho, borrowing USDC and reinvesting, inside Gauntlet's risk framework.

With infinitely many iterations at a loan-to-value ratio of L, total exposure is equity divided by (1 − L), and return on equity is:

ROE = (y − b·L) / (1 − L)

where y is the yield of the collateral and b is the borrow cost. At L = 50% the leverage is 2.0×. The equation gives the condition that settles the whole question: the loop beats simply holding the asset if and only if b < y, independent of the loan-to-value ratio. Leverage multiplies the difference between the two rates; it never changes its sign.

That matters because the yield on this kind of collateral moves. USCC, the Bitwise Crypto Carry Fund and until 1 June 2026 the Superstate Crypto Carry Fund, showed a 30-day yield of 9.18% on RWA.xyz on 4 September 2026, and about 3.56% a month earlier on the implication of the same card's 30-day change. The fund states that its 30-day return reflects the change in net asset value per share including basis accrual, staking rewards and mark-to-market effects, so it can be negative. The break-even level of a loop on such collateral moves with it.

Raise the loan-to-value ratio and the sign follows the same rule. At 70%, which is 3.33× leverage, with y = 3.56% and b = 5.5%, the loop returns −0.97%, so leverage alone turns a positive yield into a negative one. Morpho's launch of fixed-rate lending on Base in July 2026, under the name Morpho Midnight, therefore does more for this category than another collateral listing would.

Every collateral value in that equation arrives through a NAV oracle, which is the subject of chapter 6 and the reason the loop is not a purely financial calculation.

Aave Horizon reached scale, then stopped growing

Aave Horizon is a separate instance of Aave V3 on Ethereum, live since 27 August 2025, and it is the only RWA collateral market that reached measurable scale. Its design is permissioned collateral with permissionless liquidity: only an investor on the issuer's whitelist may post the RWA, while anyone at all may supply the USDC, RLUSD or GHO that gets borrowed against it. LlamaRisk recommends and maintains the risk parameters, and Aave describes the pricing layer as supplied by Chainlink through LlamaGuard Net Asset Value. Utilization, the share of deposits that has been borrowed, stood at 30.3% on the 163.5 against 539.8 mn USD reading that circulates for July 2026.

That pair of numbers is where the market's own record breaks down.

Even at its largest, the number is small against its own market: 163.5 mn USD of borrowing is 0.43% of the distributed market, and the deposits behind it are 1.41%.

Below Horizon the markets fall away by roughly an order of magnitude at each step. Morpho carries 330 to 400 mn USD of RWA collateral in mid-2026 against 240 to 270 mn USD of active borrowing, spread over 45 separate markets at the start of 2026; its architecture separates infrastructure from strategy, so a bad curator is the depositor's problem rather than a DAO's. On Solana the collateral is equities rather than Treasuries: the whole category of tokenized stocks posted as collateral peaked at 53 mn USD at the end of July 2026, with Kamino holding more than 31 mn USD of that, which is 1.3 to 2.8% of the protocol's own deposits.

Solana Compass's case study of Superstate on 16 July 2026 gives the sharpest sense of the scale: USCC had 4.4 mn USD posted as collateral on Kamino against a 140 mn USD fund, plus 44 mn USD across other DeFi markets, while Kamino's own book at the time held 2.2 bn USD of deposits and 1 bn USD of pooled loans and its main lending program saw about 140 unique wallets a day.

The largest buyer of tokenized Treasuries is a protocol's own balance sheet

Sky, formerly MakerDAO, ran a competitive process to buy the most competitive tokenized short-term US debt products. Thirty-nine teams entered, Steakhouse Financial judged them on liquidity and capital efficiency, and a governance vote split the first 1 bn USD tranche into 500 mn USD of BUIDL, 300 mn USD of USTB and 200 mn USD of JTRSY. A second 1 bn USD tranche went to the same three, taking Spark's value locked to 2.4 bn USD, which at that moment was more than two-thirds of a US tokenized-Treasury market worth 3.5 bn USD. Grove, another protocol in the same ecosystem, started separately with 1 bn USD from Sky in JAAA, the tokenized Janus Henderson AAA CLO strategy on Centrifuge. Ethena holds a further 200 to 310 mn USD of JAAA as a reserve asset as of June 2026.

Not one dollar of that roughly 3.4 bn USD is pledged against a loan. All of it is reserve. The AUM of BUIDL, USTB and JTRSY without that demand would look nothing like it does, so a large share of what gets reported as growth in the RWA market is growth in the allocation of one DAO.

A tokenized deposit and a stablecoin differ by counterparty, not by technology

The distinction most often skipped in 2026 is the one between a stablecoin and a bank deposit on a chain. A stablecoin is a liability of its issuer, covered by a segregated reserve under a reporting regime: monthly reports examined by a firm registered with the PCAOB and certified by the chief executive and chief financial officer, an annual GAAP audit above 50 bn USD in circulation under the GENIUS Act, and monthly publication of reserve composition under article 36 of MiCA. About 80% of USDC's reserve sits in a dedicated fund managed by BlackRock with custody at BNY Mellon, on a June 2026 reading.

A tokenized deposit is a liability of a bank, recorded in the bank's own books, moving on a permissioned network and governed by banking rules rather than by the stablecoin reserve regime. Kinexys, J.P. Morgan's deposit-token network, settles more than 7 bn USD a day across eight currencies on a July 2026 figure. That is a flow inside a closed system and shares no unit with the stock of distributed value quoted elsewhere in this chapter, so the two never belong in the same comparison.

The legal difference is measurable in how hard the token works. Anyone can receive a stablecoin, so its turnover is a multiple of its market value; only a whitelisted address can receive a tokenized fund. On BNB Chain, on a reading dated 17 April 2026, stablecoins worth 14.14 bn USD moved 216.42 bn USD in 30 days, about 1,531% a month, while RWA worth 3.73 bn USD moved 1.46 bn USD, about 39%. On XRP Ledger in August 2026 the same comparison runs from about 431% a month for stablecoins to about 0.45% for represented RWA. Same chain, same fees, same finality, and three orders of magnitude between them.

Most exposure to real assets on-chain involves no token at all

An RWA perpetual is a perpetual futures contract whose index is the price of a real asset, a share, an index or an ounce of gold. There is no tokenization in it, no custody and no redemption, only a price, leverage and funding paid in stablecoins. Since permissionless market deployment opened on Hyperliquid on 13 October 2025 through to May 2026, more than 100 RWA markets appeared with cumulative volume above 130 bn USD and open interest of 1.70 to 1.74 bn USD.

One figure settles the size question. Only 4.1% of RWA perpetual volume settles on tokenized contracts, on a CoinMarketCap Academy reading of around May 2026, which leaves 95.9% of on-chain turnover in exposure to real assets touching no tokenized asset whatsoever. CoinGecko's readings say the same thing from the other side: perpetuals on real assets did 524.8 bn USD of volume in the first quarter of 2026 against about 19.3 bn USD of spot value, and on tokenized shares specifically, 376.3 bn USD of perpetual volume against 7.5 bn USD of spot, a ratio of 50 to 1.

The reason is a difference in how many problems each product has to solve. Tokenization has to answer custody, the transfer agent, the whitelist, KYC, redemption and a NAV oracle before the token exists. A perpetual has to answer one question, where the price comes from. What the perpetual does not give is a claim, ownership, a dividend, a vote or a redemption right, which is why it fails all three conditions in chapter 1's test. Trading is better served by the perpetual; ownership and pledging are what tokenization is for, and mixing the two in one statistic produces most of the confusion about how big this market is.