Tokenized US Treasuries crossed roughly $7 billion in outstanding value on the last RWA.xyz snapshot I pulled. BUIDL, BENJI, OUSG, USTB, USYC, and a dozen smaller vehicles. Five separate chains market themselves as RWA-first. Together they hold less than 2% of that number.
That is the anomaly worth writing about. Three years of conference panels, nine figures of venture funding, and at least four L2s whose entire go-to-market was we are the institutional chain. The assets settled somewhere else. Ethereum mainnet holds the plurality. Stellar, of all the chains to bet on in 2021, holds a serious chunk of Franklin Templeton's BENJI. Ondo routed OUSG onto Solana over Wormhole in 2024. The RWA-branded chains got announcements.
I spent the last week reading distribution data instead of press releases. The picture is boring in the way that matters. The RWA trade is not a chain trade, and it never was. The teams building institutional L2s are solving a problem the institutions already solved with a legal opinion and a transfer agent.
Context: What RWA Actually Means Now
Start with definitions, because RWA has become a word that means anything with a legal wrapper bolted on.
Tokenized treasuries are the only RWA category with verifiable, non-speculative product-market fit. They are what they sound like: a fund or note program holding short-duration US government debt, wrapped in a legal vehicle, with a token representing a beneficial interest. The token is a share class. Nothing more, nothing less.
The category has run through three phases.
Phase one, 2021 through 2022, was DeFi-native. Maple, Centrifuge, Goldfinch, TrueFi. Undercollateralized lending to real-world borrowers, tokenized, marketed as yield. Most of it broke in the 2022 credit contraction. Goldfinch's early pool defaults were severe enough that the protocol spent most of 2023 restructuring. Centrifuge survived by rotating toward structured credit and later into treasuries through Anemoy.
Phase two, 2023 through 2024, was institutional. BlackRock's BUIDL launched March 2024 on Ethereum through Securitize. Franklin Templeton's BENJI launched in 2021 on Stellar and later Polygon — genuinely the first mover, consistently under-covered because it did not ship with a DeFi label. Ondo's OUSG began as a feeder into BlackRock's SHV ETF and now runs its own structure. Superstate's USTB, built by Robert Leshner. Hashnote's USYC, later absorbed by Circle. These are money market funds with a token interface and a compliance team behind the interface.
Phase three, 2025 onward, is collateral. The interesting question stopped being whether you could put a treasury on a chain and became whether a margin system would accept it. That is where we are, and it is where the term sheet gets real.
The plumbing is identical across every product, and it matters more than the branding:
- A legal issuer, typically a Cayman or Delaware fund or a note program.
- An administrator and a transfer agent. Securitize runs this function for BUIDL. Franklin is its own transfer agent, which is a structural advantage nobody markets.
- A NAV feed, published daily, sometimes on-chain, sometimes not.
- A permissioned token standard. ERC-3643, ERC-1400, or a bespoke allowlist contract. Transfers are gated. You cannot send BUIDL to a wallet that has not cleared an identity check.
- A cash rail. USDC, or a bank wire, or both.
Notice what is absent from that list. Blockspace. Gas. Block time. Finality assumptions. The chain is the last item and the least load-bearing one. You could rebuild the entire RWA stack on Postgres with a signed API and 95% of the users would not notice. That is not a criticism of blockchains. It is a statement about where value accrues in a compliance business.
Core: Where the Flow Actually Sits
Here is the distribution, reconstructed from public dashboards and more than a few Etherscan sessions.
BUIDL lives natively on Ethereum, plus tokenized representations on Aptos, Avalanche, Arbitrum, Optimism, and Polygon, delivered through a Wormhole-based bridge. The bridge is the interesting part. BlackRock is not operating a cross-chain message-passing protocol. Securitize is. The destination chain functions as a distribution channel for a share class, not as a settlement venue.
BENJI lives on Stellar and Polygon. Stellar, because Franklin built the original in 2021 when Stellar was the cheapest place to run a permissioned asset with a native issuer model and an existing transfer-agent integration. Polygon, because it was cheap and had institutional BD. Neither decision was made on any technical merit a protocol engineer would recognize as decisive.
OUSG lives on Ethereum, Solana, Polygon, Mantle, and Aptos. The Solana deployment arrived via Wormhole in early 2024. Again, distribution.
The cross-chain leg deserves its own paragraph, because it is the only genuinely fragile part of the stack. A bridged representation is a claim on a claim. The legal wrapper sits on Ethereum; the destination-chain token is a receipt verified by a validator set rather than by the fund's transfer agent. If that validator set is compromised, you do not lose the Treasury bill. You lose the receipt, and then you spend eighteen months in a claims process with a bankruptcy-remote entity. That asymmetry is precisely why the largest holders stay on mainnet and treat everything else as a marketing surface.
Now the arithmetic the L2s keep skipping.
Say you issue $50 million of a tokenized treasury fund. Year-one costs look roughly like this:
- Legal formation, opinion, and offering documents: $120,000 to $250,000.
- Fund administration and audit: $80,000 to $150,000.
- Transfer agent, KYC, and compliance tooling: $60,000 to $200,000 depending on holder count.
- Custody of the underlying Treasuries: 2 to 5 basis points.
- Smart contract audit, one firm, one version, one deployment: $40,000 to $90,000.
- Gas for deployment and ongoing operations: under $3,000 at 20 gwei on Ethereum mainnet, for a permissioned token with a few hundred holders and a daily NAV update.
Gas is roughly one percent of the cost of being legal. The entire L2 pitch, the one about being 100x cheaper, is an argument over a rounding error in a budget line that does not exist on the institutional P&L.
I ran a version of this against my own numbers from 2020, when I put EUR 5,000 into Curve's ETH/USDC pool and wrote a rebalancing script to test impermanent loss against emissions. Automated rebalancing beat static holding by about 14% across the high-volatility windows. The result only held after subtracting gas. At retail size, gas was 30 to 40% of the P&L. At institutional size, gas does not appear at all. Theoretical models fail without real-world gas cost consideration — that lesson is scale-dependent, and almost nobody adjusts for the scaling.
I modeled the allocation decision directly. Take $10 million. Option A: a tokenized treasury fund at 4.4% gross, 0.15% management fee, T+0 subscription in USDC, T+1 to T+2 redemption depending on the vehicle. Option B: a prime brokerage sweep into a government money market fund, 4.35% net, T+1 settlement, no gas, no wallet, no private key to custody. Option A wins by roughly 5 to 12 basis points a year before you price operational risk. It loses the moment you price the key.
That is the whole trade. Nobody wants to say it out loud because the fee revenue on Option A is somebody's business model.
Where the Yield Actually Comes From
Tokenized treasuries yield what Treasuries yield. Front-end rates sat in the 4.0% to 5.0% band through most of 2024 and 2025 depending on the curve. BUIDL, BENJI, OUSG, and USTB cluster within a few basis points of each other after fees, because they are the same underlying asset with different wrappers.
Then you look at the DeFi RWA yield products. Eight percent. Twelve percent. Fifteen percent, briefly, on a few platforms during 2024. Where does the spread originate?
There are exactly three sources, and all three are old.
Leverage. Borrow against the treasury, loop it, earn the spread between the asset yield and the borrow rate. This works until the funding rate inverts. It is a duration and funding trade with a token front end, not a new yield source.
Rehypothecation. Lend the token out, or accept it as collateral and lend against it. You are now exposed to a counterparty. If that counterparty fails, your treasury position converts into a general unsecured claim in a bankruptcy queue behind a stack of prime brokers.
Duration and credit. Buy longer paper, or non-treasury credit, and let the RWA label absorb the risk perception. The market prices the label, not the underlying.
Yield is the interest paid for patience and risk. If the advertised number is meaningfully above the risk-free rate, you have purchased risk. The only open question is whether you can name it.
I spent a winter break in 2018 tracing variable dependencies through Solidity v0.4.24 to locate an integer overflow in a price oracle feed — a bug that could have drained collateral during a flash crash. The habit that produced is simple: assume the risk lives in the code you cannot see, and go find it.
May 2022 was the same lesson at portfolio scale. Anchor advertised 19.5% on UST. The mechanism was a mint-and-burn arbitrage against a collateral pool everyone assumed was deep. It was not. I exited 48 hours before the depeg because I was watching stablecoin inflows on-chain rather than the advertised yield. The signal was structural: the yield was funded by issuance, not revenue.
Run that test on the highest RWA yield on your screen. Find the entity paying it. Find out what that entity does with the collateral. If the answer contains we lend it out, we use it for market making, or a strategic partnership with a trading firm, you have located the risk. It is not inside the Treasury bill.
The L2 Question, Answered Directly
Every L2 with an institutional strategy sells the same three things: cheap blockspace, customizable sequencing, and compliance tooling.
Cheap blockspace is priced above. It is worth less than the audit.
Customizable sequencing is a real feature that nobody has bought. A permissioned sequencer lets you censor addresses, control ordering, and offer pre-confirmations. Institutions genuinely want this. They wanted it so badly that they built it in 1995 and called it a database, and they will not migrate a transfer agent onto a rollup for the privilege.
Compliance tooling is where the L2 is least differentiated. The compliance logic does not live on the chain. ERC-3643 bakes identity into the token itself: every transfer checks an on-chain identity registry and a claim registry. ERC-1400 takes a modular route with partitions and document references. Both are EVM contracts. Both behave identically on Ethereum, Arbitrum, Base, or a permissioned Avalanche subnet. The chain contributes nothing to the compliance layer, and the compliance layer is the product.
I have argued for a while that the OP Stack versus ZK Stack comparison is not a technical question. It is a business development question. Whoever convinces more projects to deploy a chain first wins the standardization war, and technical superiority becomes a footnote in the deck. The RWA market proves it cleanly. A chain's win rate in institutional assets correlates with its BD headcount and its grants budget, not with its proving system. Franklin chose Stellar in 2021 for the issuer model and a relationship. Ondo chose Solana in 2024 for liquidity and a bridge. Neither decision involved a benchmark.
If you want to know which L2 wins RWA, skip the architecture docs. Read the partnership announcements and the foundation's grant ledger. That is the entire causal mechanism.
The AI-Agent Variable Nobody Has Priced
The real demand driver for machine-speed settlement is not tokenized treasuries. It is AI agents transacting with each other.
Agents need to pay for compute, data, and API calls at a rate no compliance department can approve. A KYC-gated permissioned token is structurally incompatible with an autonomous agent, because the agent cannot clear an allowlist check that requires a legal entity. This is not a solvable UX problem. It is a definitional one.
In 2025 I audited a payment protocol built for machine-to-machine transactions on a ZK-rollup layer. The design was elegant. The key management was not: a single hot key controlled the settlement escrow. I proposed a threshold signature scheme, a 3-of-5 with geographically distributed signers and a hardware-backed quorum, which reduced single points of failure by roughly 90%. The AI-native developers on that team had not modeled the escrow key as a target. They were optimizing throughput.
That pattern will repeat. Agents will settle in stablecoins or in purpose-built payment channels, because those are the only instruments that clear without a human in the loop. Tokenized treasuries are for balance sheets. The two markets will not merge, and anyone building an RWA rail for AI is building a bridge between a regulated fund and a bot that cannot legally hold it.
Contrarian: The Scoreboard Is Wrong
The consensus model is wrong in a specific, measurable way. The market treats RWA as a chain competition with a public scoreboard. Watch the dashboards; they rank chains by RWA value secured. That is a category error dressed up as analytics.
The scoreboard that matters is who holds the transfer agency relationship and the broker-dealer registration. Securitize holds one. Franklin holds one in-house. Ondo holds one, and acquired a broker-dealer to get there. Superstate holds one. Circle bought Hashnote for substantially the same reason. Not one of those moats is a chain. Trust the audit, verify the stack, ignore the hype — and in this sector the stack is a compliance stack, not a technical one.
Which means the L2s competing for RWA are bidding for a footer in a prospectus. The prospectus names the transfer agent, the administrator, the custodian, and the auditor. It names the blockchain in one line, and only because disclosure regimes eventually required it. The token standard is an implementation footnote.
The second contrarian point concerns collateral. Everyone is excited that tokenized treasuries now function as collateral in DeFi — in Morpho markets, in structured products, in various lending venues. Fine. But the collateral value of a tokenized treasury is only as good as its liquidity during a stress event. In March 2023, during the regional banking crisis, the market rediscovered that even real Treasuries develop a bid-ask problem when everyone sells simultaneously. A tokenized wrapper that redeems T+1 or T+2 through a transfer agent with an allowlist is not a liquid asset. It is a slow asset attached to a fast price feed. The gap between those two properties is exactly how a lending market books a bad debt hole.
There is one more thing the current cycle is hiding. The market is sideways and rate expectations are drifting lower. If front-end yields compress toward 3%, the tokenized treasury category loses its headline number. The pitch becomes 3% with a wallet instead of 3.1% in a brokerage sweep. That is a much harder sale, and the products that survive it will be the ones that were honest about their cost structure from the beginning.
Takeaway
Two things to watch, and neither is TVL.
First, the spread between the yield on the largest tokenized treasury products and effective SOFR. If that spread holds at a few basis points after fees, the products are what they claim to be. If it widens to 200 basis points without a structural explanation, someone has reintroduced duration or credit and relabeled it innovation. My alerts are set on that number, not on any dashboard.
Second, whether a tokenized treasury gets accepted as initial margin at a central counterparty, or as collateral by a prime broker, without a bespoke legal side letter. That, not a new L2, is the unlock. When it happens, the assets will flow to whichever chain the receiving institution's custodian already runs on. My base case is Ethereum, and I do not expect it to be close.
The market rewards those who read the source code. In this sector the source code is a fund prospectus and a transfer agent agreement. Read those before you read the architecture diagram.