The eight layers and who gets paid in each

Eight layers, four off the chain and four on it

The eight-layer split in the author's knowledge base is not a taxonomy for its own sake. Each layer holds different firms at a different price, and each one fails in its own way. Layer 0 is the underlying asset, layer 1 the legal wrapper that owns it, layer 2 custody and administration, layer 3 issuance and compliance. Those four sit off the chain. Layer 4 is the token contract, layer 5 the blockchain it runs on, layer 6 the data and oracle feeds, layer 7 distribution, trading and use in DeFi.

Assets under management, written AUM, is the value of what a fund holds and the base a management fee is charged on. A basis point, written bps, is one hundredth of a percentage point, so the 50 bps that BUIDL's Ethereum share class has carried since 13 November 2024 is 0.50% a year, or 5 mn USD on a billion. An issuer's economics are revenue counted in basis points against cost counted in dollars, and only the first of those scales with the size of the fund.

A tokenized Treasury fund that buys every layer from a platform, aims at 100 mn USD of assets and issues on one chain under Rule 506(c) costs about 595,000 USD in year one and about 435,000 USD a year after that. Both are estimates assembled from public 2026 benchmarks, because no RWA infrastructure vendor publishes a price list. At 100 mn USD that second figure is 44 bps a year before the issuer earns a dollar.

A fixed cost here barely moves with the size of the fund: a 10 mn USD fund and a 3 bn USD fund pay almost the same for offering documents and entity registration. A variable cost moves either with assets, like custody charged in basis points, or with activity, like gas on a transfer. Chapter 2 describes these jobs as they worked before anything was tokenized, which is what makes the comparison possible.

Custody is the bill and the holder register is the saving

Custody splits into two jobs and only one of them is old. Custody of the underlying asset answers who holds the Treasury bills; in BUIDL that is BNY Mellon, which is the fund's administrator as well. Custody of the token answers who holds the key to the wallet, and that is a separate firm on a separate contract at a separate price: Anchorage Digital Bank, BitGo Trust, Fireblocks, Copper.

The two prices are not on the same scale. A traditional fund's custody agreement charges 1.00 bps a year on the first 1 bn USD of assets and less above that; digital custody is quoted at 10 to 60 bps of assets under custody, ten to sixty times more for the same bills. The reason is amortization: a custodian bank spreads its infrastructure over trillions of dollars and decades, a digital custodian over tens of billions and a few years.

At 25 bps, the rate the knowledge base assumes for a negotiated deal and labels an estimate, digital custody is 250,000 USD of the 434,500 USD ongoing budget, or 57%. That single line explains BUIDL's 5 mn USD minimum subscription better than any regulation does, because in a model priced on assets under custody a small holder is a loss.

A transfer agent keeps the official register of who owns a fund's shares, a licensed role rather than a piece of software. Its classic price is per account: 12.00 USD per open account per year against a 25,000 USD annual minimum per fund, on a benchmark from a 1994 fund agreement, so a fund with 100,000 retail investors pays roughly 1.2 mn USD a year for the list alone. A digital transfer agent charges a flat annual fee whatever the account count, because the hundred-thousandth holder costs the same as the hundredth. This is the one layer where tokenization's advantage cannot be argued away arithmetically.

The rest of the on-chain side is less generous, because compliance costs gas. Median gasUsed for the same transfer call on Ethereum mainnet, read on 4 September 2026, was 45,160 units for USDC, 91,335 for OUSG and 183,672 for BUIDL. A BUIDL transfer costs 4.07 times what a plain stablecoin transfer costs, and the multiple counts how many compliance contracts the call has to touch.

Where a hundred basis points goes

A complete fee stack can be assembled from public sources for exactly one product. An OUSG investor pays about 35 bps all in: 15 bps to Ondo as a capped management fee, and 20 bps to BlackRock through BUIDL in a reduced-rate share class. How that 20 bps is divided among Securitize, BNY Mellon, PwC and BlackRock itself is disclosed by none of the four, which makes it the largest undisclosed number in this market's economics and the reason the rest of this section is a model of a 500 mn USD fund in a European or offshore wrapper trading on an ATS at an all-in 100 bps. Every line is anchored to a published market rate, and the whole thing is an estimate rather than an invoice.

Two readings come out of that table. The manager takes the largest share and not a majority, at 35 bps, which is close to the proportion it takes in a fund that was never tokenized. What changed is the arrival of a layer a classic fund does not have at all, the platform and the transfer agent at 12 bps, and that is the direct price of being on a chain. Distribution at 15 bps is the second-largest line: tokenization did not remove the intermediaries, it moved them from advisor networks to whoever controls the allow-list and the exchange integrations.

The issuer's net margin is what is left when every other layer has been paid, 5 bps in this model, and on a 500 mn USD fund that is 250,000 USD a year. At the 15 to 20 bps the Treasury market actually charges, the same arithmetic turns negative.

Break-even, and why the usual division gets it wrong

Break-even AUM is the level of assets at which the fee a fund collects equals what it costs to run. The knowledge base computes it as one division, fixed annual cost over the net fee rate, and its estimates are about 87 mn USD at a 50 bps management fee, the rate reported for BUIDL on Ethereum, and about 290 mn USD at the 15 bps estimated for OUSG and USTB. Building rather than buying moves the numerator, and the same source offers two estimates of that move which should not be averaged: an added 960,000 USD in year one and 835,000 USD a year after puts the estimated threshold at 15 bps near 847 mn USD, while a wider estimate of 2.5 mn USD a year puts it at 1.67 bn USD.

A management fee is a percentage of assets accrued daily and deducted from NAV, so the investor never sees a payment leave an account, only a lower yield. The net rate does the work in that formula: an issuer charging 15 bps on a product whose portfolio sits inside BUIDL runs on 15 while its investor pays about 35 all in.

Recomputing those thresholds from the cost lines the same source publishes does not reproduce them, and the reason is one line item. A quarter of a basis point structure means most of the cost moves with the assets: digital custody at 25 bps and asset custody at 1 bps are 260,000 USD of the 434,500 USD budget, and both rise one for one with the quantity the division is solving for. A cost that scales with assets cannot sit in the numerator of that division. Splitting the same budget instead gives 174,500 USD that stays put and 26 bps that does not.

On that split the author's own estimated threshold at 50 bps is about 73 mn USD, below the knowledge base's 87 mn USD, and at 15 bps there is no threshold at any level of assets, because a 15 bps fee never overtakes a 26 bps custody bill. The narrow build estimate moves the fixed part to about 1.01 mn USD and the estimated 50 bps threshold to about 421 mn USD, and leaves 15 bps with no solution. The wider 2.5 mn USD estimate cannot be resplit, because the source gives it as one annual number without saying how much of it scales.

Real products fall on both sides of that line. BUIDL at 50 bps on 2.8 bn USD of assets in July 2026 generates about 14.0 mn USD a year, at a marginal cost close to zero for a firm that runs Treasury infrastructure anyway. OUSG, with its waiver in force, generates nothing for Ondo. Two kinds of firm clear the threshold: those for whom a tokenized fund is a new channel for a business they already have, and those for whom the fund feeds something else, as USYC feeds circulation in USDC.

Scarcity is not in the software. RWA.xyz counted 207 tokenization platforms around September 2026, 196 distributed and 23 represented, so nobody is short of a vendor. What is short is registrations: Securitize runs a regulated US offering through four entities holding four separate statuses, broker-dealer and ATS operator, registered transfer agent, exempt reporting adviser, fund administrator. Money does not buy those quickly.

The price the investor sees is paid out of three other pockets

The rates on display today are not equilibrium rates. Three pockets cover the gap between them and what the product costs to make, and all three are temporary by construction. The first is blockchain foundations. BUIDL's management fee depends on the network: 50 bps on Ethereum, Arbitrum and Optimism, and 20 bps on Aptos, Avalanche and Polygon. BlackRock does not absorb the difference. Aptos Foundation, Avalanche (BVI), Inc. and Polygon Labs BD Investments (Cayman) Ltd. agreed to pay BlackRock a quarterly fee so that the rate on their chains would be the lower one, announced on 13 November 2024. A foundation is buying the presence of a BlackRock product in its network, and the investor gets 30 bps off until the agreement ends.

The second is fee waivers. A gross expense ratio is what a fund's operation costs before anything is given back; a net expense ratio is what the investor pays after caps and waivers. A waiver is a temporary suspension of a fee to build assets, a campaign rather than a price. Paxos suspended its PAXG creation fees until 31 March 2026, and OUSG produces zero management-fee revenue for Ondo because the whole yield goes to NAV. FOBXX shows how easily the three numbers that all get called a fee are confused.

The comparable figure across products is the gross one, and the number quoted in marketing is the net one.

The third pocket is shareholder and venture capital, spent as operating loss. It is the most used source in this sector and the least often named. Securitize lost 7.93 mn USD in the first quarter of 2026 and 21.7 mn USD in the second, close to 30 mn USD in half a year at one company, reported in its 8-K and 8-K/A filings of 12 August 2026.

The oracle layer is subsidized in a way that leaves no trace in the fee table. Issuers generally do not pay for NAV feeds directly: chains absorb the cost through programs like Chainlink SCALE, and DeFi protocols absorb it when they want an asset as collateral. Neither Chainlink nor RedStone publishes a price for NAVLink, Proof of Reserve or their Securitize feeds, so the 3 bps line is the least anchored number in the model. Chapter 6 reads those feeds contract by contract.

Four of six layers sit above a 70% threshold

Competition at the platform layer is real, and concentration further along the chain is not. Ten assets accounted for 76.6% of tokenized-stock trading volume in CoinGecko's September 2026 report, and the same report put more than 70% of open interest on two venues. About 60% of RWA volume on centralized exchanges runs through a single venue on Stobox's count. Two tokens, XAUT and PAXG, are 71% to 90% of the tokenized gold category depending on which of two sources is read.

On the credit side the picture is sharper still. JTRSY and JAAA, both managed by Janus Henderson, are 95.6% of Centrifuge's total value locked, and more than half of Grove comes from one allocator, Sky. Centrifuge's 5.21 mn USD of annual protocol revenue therefore rests on two clients. Chapter 9 profiles the firms holding those shares.

The one issuer that files shows where the model breaks

Securitize is the only pure-play tokenization company with public reporting obligations, which makes it the only place where this layer can be read from a filing rather than a deck. It reports two lines: tokenization, project revenue for bringing an asset on-chain, and asset servicing, recurring revenue from transfer agency, fund administration and reporting.

In the second quarter transaction volume rose 147% quarter over quarter, to 5.3 bn USD, while revenue fell 5% year over year. The model charges for launching an issue and for the stock of assets under service, and nothing on turnover, so rising activity costs the platform money instead of earning it.

The most profitable models in this market sell something else entirely. Figure Technology Solutions originates home equity loans, records them natively on the Provenance blockchain and sells them on. Its first-quarter 2026 net revenue was 119.79 mn USD and its 10-Q contains no line called tokenization revenue. The benefit of the chain shows up instead as a lower cost of capital: AAA tranche spreads in Figure's home equity securitizations compressed from 255 bps in 2023 to about 135 bps year-to-date in 2026, across 22 priced transactions and more than 100 buyers. Those 120 bps are the only benefit of tokenization in this market expressed in basis points and tied to a named mechanism, a native on-chain loan registry giving buyers better data on the collateral, and Figure and its partners keep them.

A take rate is the share of the fees flowing through a protocol that the protocol keeps, as distinct from the fees themselves, which mostly belong to somebody else. Maple Finance ran a take rate of 13.0% in DefiLlama's September 2026 reading, keeping 14.43 mn USD out of 111.07 mn USD of annualized fees. Its revenue is interest margin and origination commission, a function of turnover rather than of deployments closed in a quarter, and that is the property Securitize's model lacks.