Investing in Tokenized Real-World Assets

On Saturday, April 13, 2024, an ounce of gold cost either $2,449.14 or $2,361.69, depending on which token you asked. PAX Gold and Tether Gold are the two large tokenised gold coins. Each is a claim on one fine troy ounce of allocated bullion in a vault, each is audited, each is redeemable, and on that Saturday they were 3.7% apart on Coin Metrics’ reference prices, the widest gap in the 874 days we measured from January 1, 2024 to May 23, 2026. Nothing had happened to the gold. The professional bullion market was shut for the weekend, and the two tokens were left to be priced by whoever wanted them at four in the morning on a crypto exchange.

That gap is the whole subject in one number. Tokenisation does not create an asset, improve one, or change what one returns. It creates a claim on an asset that settles in seconds, on a public ledger, at any hour, in any jurisdiction with a phone. Everything good about a tokenised asset comes from that. Everything bad comes from the same place: a claim that trades continuously against an underlying that does not, held through a chain of legal entities most buyers never read, and worth what it is worth only while somebody stands behind redeeming it.

The market that resulted is not the one the pitch described. The technology that was sold in 2021 as a way to own a fraction of a Picasso is, in 2026, mostly a money-market fund. BlackRock’s BUIDL, launched with Securitize in March 2024, passed $1B on March 13, 2025 and stood at about $2.8B in early September 2026, on rwa.xyz’s count as reported. For scale: the combined floor market capitalisation of the four blue-chip NFT collections our tape tracks, CryptoPunks, Bored Apes, Pudgy Penguins and Azuki, was about $1.01B on September 8, 2026. One institutional cash fund is worth nearly three times the entire blue-chip JPEG market it was built on top of. The Investing in NFTs guide covers that market, and three sister guides take its surviving corners — Investing in CryptoPunks the deepest collection, Investing in Generative Art the segment the museums kept, Investing in Bitcoin Ordinals the same idea rebuilt on another chain. This one is about the plumbing that outlived all of them.

A token is a row in a register that happens to live on a public blockchain, and the first job is to work out which register it is a row in.

Take the ordinary version first. When you own a share of a money-market fund, your name sits in a book kept by a transfer agent, a regulated firm whose function is to record who owns what and pay out accordingly. Tokenisation moves that book onto a blockchain. The token in your wallet is the entry; moving the token moves the entry; the transfer agent recognises the chain as the authoritative record. Nothing about the fund changes: same Treasury bills, same prospectus, same custodian, same board. What changes is that settlement takes seconds instead of a day and works on a Sunday.

That is one of three architectures, and the differences decide what you actually hold when something goes wrong.

The register model

The token is the ownership record for a security that exists in law. Tokenised money-market funds work this way. The issuer is a real fund, the register is on-chain, and transfers are restricted to wallets the transfer agent has approved. You own the fund share. Your recourse in a failure is the fund’s own legal structure: custody rules, segregation of assets, the ordinary machinery of securities law.

The warehouse-receipt model

The token is a bearer claim on a specific physical thing sitting in a specific vault, redeemable by the holder. Tokenised gold and tokenised graded trading cards work this way. You do not own a share in a company that owns gold; you own a right to demand delivery of an ounce. Your recourse in a failure is the vault, the insurance policy over it, and whatever the issuer’s terms actually promise, which is often less than the marketing does.

The wrapper model

The token is issued by a special-purpose vehicle that holds the asset, and you own an interest in the vehicle, or a debt claim against it. Tokenised equities, tokenised real estate and most tokenised private credit work this way. You do not own the share, the building or the loan. You own a claim on an entity that owns it, which means you have taken on the credit of that entity, its administrator, and the jurisdiction it was formed in, on top of the asset risk you thought you were buying.

The three are not interchangeable and the marketing language is identical across all of them: “backed by,” “redeemable for” and “represents” do very different legal work in each case. The single most useful habit in this market is to ask, of any tokenised asset, which model it is, who is obliged to give you the underlying, and whether anyone has ever made them do it.

A tokenised asset returns what the asset returns, minus the wrapper, and the honest record is a subtraction problem rather than a performance table.

There is no index of tokenised real-world assets with a long history, because the category is too young and too heterogeneous to have one. What exists instead is a set of arithmetic identities you can check for yourself:

• A tokenised Treasury fund returns the bill yield minus the fund’s management fee minus on-chain transaction costs. Our survey of the accessible funds in May 2026 put the fee drag in a range of roughly 15 to 50 basis points. With the Federal Reserve’s target range at 3.50% to 3.75% after the July 28–29, 2026 meeting, that is a return in the low threes, and it is fully taxable as ordinary income.
• A tokenised gold coin returns the gold price minus what the issuer takes at the door — neither of the two large ones charges an annual storage fee at all, which is their genuine edge over a bullion ETF — and gold’s own record is the Investing in Gold guide’s subject. Our tape had gold at $4,443.90 an ounce on September 8, 2026 on Yahoo’s futures series.
• A tokenised equity returns the share price and the dividend net of 30% US withholding, reinvested rather than paid, but not the vote and not the shareholder’s legal position.
• A tokenised collectible returns what the card, the watch or the bottle returns, minus the marketplace commission, minus the cost of getting the physical object back. Storage, on the two card platforms, is free.

That is the entire performance story, and it is why the marketing so rarely shows returns. The pitch has never actually been about return. It is about access, granularity, settlement in seconds and trading at any hour. Each is real, and each has a price paid in liquidity and legal certainty rather than in an explicit fee, which is what makes it easy to miss.

Compare it honestly with the alternative. An S&P 500 index fund at three to four basis points gives you the equity risk premium, a shareholder’s legal position and the long-term capital-gains rate. A tokenised money-market fund gives you a bill yield taxed as ordinary income, a wrapper fee on top, and a settlement improvement you probably do not need unless you run a trading desk. The case for tokenisation is strong for people already on-chain and weak for people who are not, and no amount of technology changes that.

Nearly half the tracked market is one product sold to one kind of buyer, and seeing that clearly is the fastest way to calibrate everything else.

rwa.xyz is the standard tracker and its headline number excludes stablecoins, which are a separate and much larger market covered in the Investing in Stablecoins and On-Chain Yield guide. On its count as reported, tokenised real-world assets outside stablecoins were about $31.4B in mid-May 2026, read $35.2B on July 10, 2026 and stood near $34.7B in August 2026. Tokenised US Treasury funds were about $13.4B in early April 2026 and $16.2B on August 3, 2026 on rwa.xyz’s own treasuries page, which is to say that a little under half the entire category is one product: a money-market fund with a blockchain register.

Treat the total as an order of magnitude rather than a reading. Trackers count differently and disagree by billions: through July and August 2026 the same market was published at $29.5B by Allium (July 15, on-chain crosschain data), at $33.5B by rwa.xyz in early July, and at $37.9B on an rwa.xyz “distributed asset value” measure on August 6. Anything quoted above $60B is counting stablecoins or idle collateral. The shape of the market is not in dispute; its second decimal place is.

The residual matters for what it says about the rest. Take mid-May 2026, where both figures come from the same source and month: a $31.4B total against roughly $15B of Treasury funds leaves about $16.4B of everything else, and on our own computation the two large tokenised gold coins alone were $5.3B of it on May 23. Private credit, tokenised equities, real estate and commodities other than gold share what is left. None of them is a large market by the standards of the thing it tokenises.

Set that against the market next door. Dollar stablecoins stood at $301.7B on September 3, 2026 on public trackers, 6% below their record $321.2B of May 24, 2026. The yield-bearing, fund-registered, legally cleaner version of an on-chain dollar is therefore about a twentieth the size of the version that pays you nothing. That comparison is the most informative number in the category, and it tells you the market is buying settlement convenience rather than yield: people who want a dollar on a blockchain overwhelmingly take the one that is easiest to move and hold, not the one that pays them interest.

The largest and best-built corner of the market is also the one most readers cannot buy, and the reason is a minimum, not a technology.

The product is straightforward. A fund holds short-dated US government debt. Its shares are recorded on a public blockchain instead of, or alongside, a conventional transfer agent’s book. Income accrues daily and is paid in new tokens or in a rising token price.

The names to know are BlackRock’s BUIDL, run with Securitize, launched in March 2024 and about $2.8B in early September 2026; Franklin Templeton’s BENJI, the on-chain share class of its US government money fund FOBXX, which has run on Stellar since April 2021 as the first US-registered mutual fund to use a public blockchain as its official record of share ownership and is the oldest of them; Circle’s USYC, acquired with Hashnote in January 2025 and the largest of the group since it passed BUIDL in mid-March 2026; and Ondo Finance’s OUSG and USDY. On the survey behind our stablecoin guide, dated May 2026, BUIDL was about $2.9B, USYC about $3B, BENJI about $2.3B, USDY about $740M and OUSG about $670M, in a category then crossing roughly $15B.

Three structural facts govern whether any of this is available to you.

They are securities, and the register enforces it

A tokenised fund share moves only between wallets the transfer agent has whitelisted; a transfer to an unapproved address reverts. That is the mechanism by which a regulated fund can live on a permissionless chain at all, and it also means you cannot post the token as collateral wherever you like. The “composability” tokenisation is sold on stops at the whitelist.

The minimums are institutional

BUIDL is restricted to qualified purchasers at a $5 million minimum subscription. USDY is offered under Regulation S to non-US persons, at a $500 primary-market minimum; its US-eligible sibling OUSG requires qualified-purchaser status, which under section 2(a)(51) of the Investment Company Act means $5 million of investments for an individual and $25 million for an institution. That is the pattern: the large tokenised Treasury funds are sold to crypto exchanges, market makers, DAOs and corporate treasuries that need a yield-bearing asset settling at crypto speed. Franklin Templeton’s Benji is the notable exception, a retail-accessible on-chain share class of a registered fund, and worth watching precisely because it is the counterexample.

Redemption is a business process, not a market trade

You do not sell a tokenised fund share on an exchange; you present it to the issuer, who wires you dollars or mints you a stablecoin on the fund’s own schedule, often near-instantly into a stablecoin during market hours and T+1 otherwise. The instant leg is worth naming, because it is the clearest example of what you are actually relying on: since April 11, 2024, BUIDL holders have had a Circle smart contract holding a pool of USDC that exchanges BUIDL for USDC one-for-one, around the clock. That is not the fund redeeming; that is a third party standing in front of the fund with its own inventory. An “instant” redemption is a liquidity facility somebody is providing, and liquidity facilities are withdrawn exactly when you want them.

Where that yield goes once it is on a chain is the subject of the Investing in Stablecoins and On-Chain Yield guide; this one takes the fund as an asset rather than as a yield source. The point worth restating is the tax one, because it is the most common error: a tokenised money-market fund pays interest, interest is ordinary income, and you owe it in the year it accrues whether or not you have sold the token.

Tokenised gold is the cleanest test of the whole idea, because the underlying is fungible, universally priced and sitting in a vault, and the record shows exactly what the wrapper adds and costs.

PAX Gold (PAXG), issued by Paxos, and Tether Gold (XAUT), issued by TG Commodities, each represent one fine troy ounce of allocated London Good Delivery gold. Allocated matters: specific bars, serial numbers published, held in the holder’s beneficial interest rather than on the issuer’s balance sheet. Neither charges an annual storage or custody fee, which is a real advantage over a gold ETF: the largest, SPDR Gold Shares, charges 0.40% a year, and even the cheap mini version charges 0.10%.

Both issuers earn at the door instead. Paxos charges a creation and destruction fee that runs from 100 basis points down to 12.5bp on large orders, plus 0.02% on an on-chain PAXG transfer; Tether charges a flat 25bp when a verified customer buys or redeems through TG Commodities, whatever the size. What the wrapper does not change is the tax: section 408(m) of the Internal Revenue Code lists “any metal or gem” among collectibles, so a tokenised ounce is taxed the way bullion and the bullion ETFs are, at up to 28% on a long-term gain rather than 20%. The tax section below works that through.

Redemption is the term to read, and here the marketing and the mechanics diverge sharply. Both tokens require roughly 430 of them to take delivery of physical metal, because a London Good Delivery bar runs between 370 and 430 fine ounces and neither issuer will cut one. At the $4,443.90 gold price on our tape for September 8, 2026, that is about $1.9 million of metal to exercise a right the marketing describes as though it were a coin shop. Paxos delivers the bar to a vault in the United Kingdom; Tether arranges secure transit of the bar to a Swiss address, or will sell it into the Swiss market and pay you the proceeds.

The escape hatch is a partner: Paxos’s Alpha Bullion redeems PAXG for physical gold in sizes from 1 gram to 1 kilogram, which is the only retail-scale physical redemption either coin offers. Everyone else redeems the way everyone actually does — by selling the token for dollars, at the creation-and-destruction fee above or at whatever the exchange charges.

On our own computation from Coin Metrics Community Data, the two coins held 1,178,048 ounces, about 36.6 tonnes and $5.3B, on May 23, 2026, against a combined $167M at the end of 2020, a 32-fold increase.

But the shape of that growth is more interesting than its size. Through 2023 and 2024, gold rose 44.5% in dollars on Coin Metrics’ own token prices, from $1,818 an ounce at the end of 2022 to $2,628 at the end of 2024. Tokenised ounces outstanding fell 13.7% over the same two years, from 516,910 to 446,081. Holders redeemed into a rising market. That is what a working warehouse receipt looks like: the token is a way to move and pledge gold, not a way to be long gold, and when the price rose the people using it as collateral took delivery. The supply then doubled through 2025 and into 2026 as the gold price went to record levels, which is the ordinary retail pattern arriving late.

Now the cost the wrapper imposes, which no fee schedule discloses. Gold’s professional market runs on London and Comex hours. A tokenised ounce trades every minute of every day. Between the close of one and the open of the other there is no arbitrage against the physical market, only against other crypto buyers, so the token’s price is a crypto-market opinion of gold rather than gold. We measured this directly by comparing the two tokens with each other, which removes any argument about which gold benchmark to use: two claims on the same ounce, from two issuers, on the same day.

Over the 874 days from January 1, 2024 to May 23, 2026, PAXG and XAUT closed more than 0.5% apart on 290 days, a third of the sample, more than 1% apart on 86 days, and more than 2% apart on eight. The widest, 3.7%, was the Saturday in April 2024 that opens this guide. The median absolute gap was about a third of one percent. That is the real transaction cost of a tokenised commodity, and it is paid by whoever transacts at the wrong hour, not disclosed anywhere.

Tokenised shares are the most heavily marketed corner of the category and the one where the gap between the token and the thing is widest, in law and in liquidity.

The two visible venue families are Backed’s xStocks and Ondo Global Markets, renamed Ondo Stocks on July 13, 2026 with no change to the structure beneath it. An xStock is a bearer tracker certificate issued by Backed Assets (JE) Limited: a ledger-based security under article 973d of the Swiss Code of Obligations, sold on a base prospectus filed with Liechtenstein’s financial regulator and collateralised by shares in a custody account governed by a three-party control agreement with an independent security agent, who can liquidate the collateral and pay out token holders if the issuer fails. An Ondo token is issued by Ondo Global Markets (BVI) Limited against shares bought and custodied through Alpaca Markets, a US-registered self-clearing broker-dealer, in control accounts at the Depository Trust Company.

Both give price exposure. Neither, in the structures on offer through 2026, gives you what a shareholder has: you are not on the issuer’s register, you do not vote, and no proxy is passed through. The dividend is the one place the wrapper does more than nothing, and it is worth knowing exactly what it does. Backed does not pay the dividend to you; it reinvests it in more of the underlying and raises a multiplier on your token, so a payment worth a tenth of a share moves your balance from 1.0 to 1.1 — after 30% US withholding tax has been deducted, a permanent leak that a US taxpayer holding the actual share would not suffer. You have swapped equity risk for equity risk plus the credit of a certificate issuer.

The liquidity argument is worse than the legal one. We took Coin Metrics’ reported spot volume for seven tokenised equity wrappers and computed the median day. Over the 298 days from July 30, 2025 to May 23, 2026, the tokenised Circle share (CRCLx) was the busiest of them at a median $2.4M a day, with a single best day of $25.2M. The tokenised Coinbase share traded a median $833K a day on the Backed wrapper and $469K on the Ondo one; the tokenised MicroStrategy share managed $77K a day on Backed. These are the flagship names of the category, and a median day in the busiest of them would be a rounding error in the listed share.

The category has grown since our window closed — Ondo’s platform passed $1B of total value locked in July 2026, less than eight months from launch, and claims more than 70% of the tokenised-equity issuer market — but locked value is not traded volume, and it is the traded volume that sets your exit price.

The data does say one thing in their favour: arbitrage between wrappers works. On the seven overlapping days in May 2026 for which Coin Metrics published reference rates for both families, the median gap between the two tokenised versions of the same share was under two-tenths of one percent — though the full range across those seven days ran from −0.97% to +0.66%, so single days are wider than the median suggests. Seven days is a very small sample; Coin Metrics does not publish reference rates for these wrappers before May 18, 2026. The plumbing appears to function; the question is what it is for.

Where the underlying is illiquid, tokenisation removes none of the illiquidity and adds a layer of intermediaries, which is why this end of the market has the worst record and the loudest marketing.

Private credit is the largest non-Treasury category on rwa.xyz’s breakdown, with more than $14B of active loans reported in 2026. The structure is almost always the wrapper model: an originator makes loans, a special-purpose vehicle holds them, and a token represents a share of the vehicle’s cash flows. Platforms including Figure, Centrifuge, Maple and Goldfinch have run versions of this since 2020, and the honest summary is that on-chain private credit has repeated the offline lesson faster.

Two defaults are worth carrying in your head, because they are the record. In December 2022, Orthogonal Trading defaulted on $36M across eight loans on Maple Finance after telling its lenders in November that its exposure to the collapsed FTX was about $2.5M and then disclosing on December 3 that far more capital was trapped there. Roughly $31M of that sat in a single USDC pool run by the credit manager M11, and the remaining lenders in that pool took a loss of about 80%; a smaller WETH pool lost about 17%.

In 2023, Tugende Kenya, a motorcycle-taxi financier, became Goldfinch’s first borrower default on a $5M USDC loan after breaching covenants by on-lending $1.9M to a related entity; the loan was written down by $3.1M in stages, a net hit of about 3.95% to the senior pool’s NAV, and the protocol’s legacy pool was wound down. Both losses ran the same way an off-chain default runs: slowly, through lawyers, with no help from the chain.

The mechanism needs no further proof: a token cannot make an unsecured loan to an emerging-market lender into a liquid asset, and the 8% to 12% being quoted is that loan’s credit spread, not a technology dividend. Note also which platforms survived and how: Maple’s post-2022 answer was to move to overcollateralised lending, which is to say it stopped doing the thing that made the yield interesting. The Investing in Private Credit and Investing in Listed BDCs guides cover what that spread has paid in regulated form.

Tokenised real estate is the version most readers have been pitched, and on July 2, 2026 it produced the category’s clearest failure so far. Two platforms carried the US retail market: RealT and Lofty. Both sold membership interests in a limited liability company owning a single property, in small denominations, and both put the rent on-chain. One of them is now being wound up.

RealT withdrew from US investors in 2023. In July 2025 it was named in what Detroit officials called the largest nuisance-abatement suit in the city’s history, covering 408 properties and 165 affiliated corporate defendants. Rent distributions stopped in February 2026. A court-appointed fiduciary took over in April. On July 2, 2026 the company announced a voluntary liquidation of everything it held, including more than 700 Detroit houses. Reported estimates put roughly $640,000 in the investor escrow against about $140M raised from something like 14,000 investors: on the order of $45 a head.

Read that against this section’s claim and it fits exactly. The tokens stayed transferable the whole way down. What became unsaleable was a portfolio of Detroit houses with a code-enforcement problem, and no amount of settlement speed touched it. A tradable claim on an illiquid asset is a tradable claim, not a liquid asset, and RealT’s version of the argument ended in a fiduciary’s hands.

Lofty is the survivor, and its structure is worth stating precisely, because it is routinely described wrongly. Each property sits in its own Wyoming LLC. The token is an Algorand Standard Asset representing a membership interest in that LLC. The retail minimum is $50. As of September 2026 the platform reports more than 180 properties across 21 states, about $100.5M of cumulative invested value and over 40,000 investors.

The regulatory position is the part to read twice. Lofty makes no Regulation D, Regulation CF or Regulation A+ filing for the retail marketplace, relying instead on a Wyoming intrastate LLC and a direct-ownership theory that no regulator or court has tested. That is not a private-placement exemption; it is an argument that no exemption is needed. Price it accordingly. What you own on either platform is an LLC interest, not a deed. What you receive is your share of rent after a property manager takes a fee. What you cannot do is sell easily: secondary liquidity sits inside the platform’s own app rather than on a public exchange, and the discount to the last valuation can be severe, because the valuation is the sponsor’s and the bid is the market’s.

Tokenisation also cannot fix the part of real estate that actually goes wrong. A tenant stops paying. The roof needs $18,000. The city files a code violation. Each is handled by a property manager you did not choose, whose incentive is the management fee, in a market you have never visited, and your recourse is an operating agreement governed by another state’s law. The Real Estate Syndications and Private REITs guide and the Investing in Short-Term Rentals guide cover the same economics without the token. If a deal is good, tokenisation makes it slightly more tradable. If a deal is bad, tokenisation makes it faster to sell you.

Tokenisation earns its keep at the bottom of the collectibles market, and the reason is entirely about the cost of moving small physical objects.

Courtyard, on Polygon, and Collector Crypt, on Solana, do the same thing: take a graded trading card, put it in a professional vault, mint one token per card, and let the token trade or be redeemed for the physical card. Courtyard vaults with Brink’s and, on CryptoSlam’s 2025 rankings, led the entire top-ten NFT table on transaction count, with weeks above 100,000 sales; it was the engine that carried Polygon’s NFT volume past Ethereum’s. Collector Crypt reports more than $1.6B of lifetime volume and roughly two-thirds of the tokenised-card market on Solana.

The economics work because the friction they remove is unusually large relative to the asset. Selling an $80 card conventionally means listing it, packing it, insuring it, shipping it, and absorbing a return if the buyer disputes the condition. eBay alone takes a 13.25% final-value fee on trading cards plus $0.40 an order, and shipping, insurance and supplies push the all-in cost toward a fifth of the sale price. Selling a vaulted card as a token costs a marketplace fee and a fraction of a cent of network fee, and the buyer takes possession of a graded, sealed, photographed object that never moves.

The published schedules are unusually cheap, which is the finding rather than an aside. Courtyard charges no marketplace fee at all on ordinary listings, and has not since June 3, 2025; it takes a 10% buyer’s premium on auctions and, during heavy demand, a $2-a-card handling charge on redemption, plus shipping. Collector Crypt takes 2% out of the seller’s proceeds on a sale, charges 2% of the card’s insured value as a vault-withdrawal fee when you redeem, plus shipping and handling, and charges nothing at all for storage or insurance while the card sits there; delivery runs two to three weeks domestically and about five internationally.

Collector Crypt also runs a standing on-chain buyback at 85% to 90% of an indexed value drawn from eBay and other markets, which is the real bid under the market and a 10% to 15% haircut, not a 2% one.

Our tape follows Collector Crypt’s order book directly. Between August 28 and September 8, 2026 the number of cards listed rose from 130,614 to 142,884, up 9.4% in eleven days, while the median asking price fell from $90.00 to $85.98, down 4.5%. Read that second number carefully: over the same ten readings the median ask ranged from $75.00 to $115.00, so 4.5% is the distance between two endpoints rather than a trend. Multiplying a median by a count is not a market capitalisation, but as an order of magnitude it puts the entire listed book somewhere around $12M: this is a market of ordinary cards, not grails. For comparison, our physical card basket on PriceCharting had a median ungraded price of $295 and a median PSA 10 price of $5,848.66 on September 8, 2026, both several times the tokenised median.

That gap is the finding, and a durable one. Tokenisation is winning at the bottom of the collectibles market, where handling costs dominate value, and losing at the top, where a serious buyer wants the object in hand and a couple of points of marketplace fee is trivial against a six-figure hammer. The Investing in Sports Cards, Card Grading Explained and Pokemon TCG guides analyse the underlying market; the token is plumbing, and good plumbing for cheap things.

Every tokenised asset carries powers that a share certificate does not, and the terms that describe them are the most important pages in the offering document.

Start with transfer restrictions. A tokenised security enforces investor eligibility in code: a whitelist of approved wallets, checked on every transfer. The token cannot reach an unapproved venue, cannot be lent into a protocol that is not whitelisted, and cannot be sold to a buyer who has not completed the same identity checks you did. Anyone promising a liquid secondary market for a whitelisted security token is describing a market the whitelist itself prevents.

Then the administrative powers. Most tokenised-asset contracts include a function to freeze a balance, force a transfer, or burn and reissue tokens. Those exist for good reasons, a lost key, a court order, a sanctions listing, and they are the same powers a transfer agent has always had over a book-entry security. They are also why “self-custody” means something different here: you hold the key, and the issuer holds the ability to make your holding unspendable. Both are true at once.

Then bankruptcy. Ask where the asset sits if the issuer fails. In the register model, the fund’s assets are the fund’s, held by a custodian, and the issuer’s insolvency is a servicing problem rather than a loss of principal. In the warehouse-receipt model, the answer depends entirely on whether the bullion or the cards are held in trust for token holders or on the issuer’s own balance sheet; allocated, segregated, in-trust custody is the language that matters, and it should appear in the terms, not only in the marketing. In the wrapper model, you are an unsecured creditor of, or an equity holder in, a special-purpose vehicle, and you should read the vehicle’s jurisdiction the way you would read a bond’s covenants.

Finally, the oracle. Where a tokenised asset is used as collateral, some system has to tell the lending contract what it is worth. On November 4, 2025, Stream Finance disclosed a $93M loss and its yield-bearing token xUSD fell about 77% to $0.26 within a day, and has since traded between seven and fourteen cents. Vaults on Morpho, Euler and Silo went on pricing it at a hard-coded $1.00 — on some Euler markets, at $1.27 — so the liquidations that should have unwound the leveraged loans never fired, and borrowers walked off with the supplied dollars instead. The lesson generalises past stablecoins: a tokenised asset’s on-chain price is an assertion by whoever wrote the price feed, and it can be wrong for as long as nobody updates it.

US policy moved from enforcement to rule-writing between 2025 and 2026, and the practical question for a buyer is unchanged by any of it.

The sequence explains why tokenised products became available at all. The SEC’s enforcement cases against the major exchanges were dismissed in the first half of 2025. On July 31, 2025 Chairman Paul Atkins announced “Project Crypto,” a programme to rewrite the rules so securities can be issued and traded on public chains.

On March 17, 2026 the Commission issued a 68-page interpretive release, which the CFTC joined, setting out how the Howey test applies to crypto assets, with a five-part taxonomy separating digital commodities, digital collectibles, digital tools, stablecoins and digital securities — only the last of which is inherently a security. On August 18, 2026 it proposed Regulation Crypto Assets (published in the Federal Register on August 21): a start-up exemption of up to $5M over four years, a fundraising exemption of up to $75M in any twelve months, both carrying narrative disclosure obligations, and a conditional safe harbour from investment-contract treatment. As of September 2026 that is a proposal, not a final rule.

Congress has moved more slowly. The GENIUS Act, signed July 18, 2025, governs payment stablecoins and binds by January 18, 2027 at the latest; it is the reason a tokenised dollar and a tokenised money-market fund are now clearly different regulated products, and it bans issuers from paying interest on the former.

The CLARITY Act, the market-structure bill that would define when a token is a commodity, passed the House 294–134 on July 17, 2025 and was reported out of Senate Banking 15–9 on May 14, 2026, with a cloture vote set for 2:15pm Eastern on September 15, 2026. That vote is a 60-vote procedural test on the motion to proceed, not final passage; even a yes leaves the text, the amendments, the floor vote and reconciliation with the House ahead. The industry has been “weeks away” from market-structure law since 2023; treat it as pending until it is signed.

Banking regulation is the quieter half and probably the more consequential. National trust charters were conditionally approved for Circle, Ripple, Paxos, BitGo and Fidelity Digital Assets on December 12, 2025, joining Anchorage, chartered in 2021. A nationally chartered trust company is a custodian a fiduciary can use — it may hold client assets but may not take deposits or lend — and custody, not blockchain throughput, is the binding constraint on institutional tokenisation.

What none of this changes is the analysis you have to do. A regulated wrapper is a better wrapper; it is not a better asset, and it does not remove the four questions that decide the outcome: who holds the underlying, who is obliged to give it to you, what it costs to make them, and what happens in their insolvency. A tokenised fund that is fully compliant with a final Regulation Crypto Assets will still be a bill fund with a fee.

The cost stack has five layers, most buyers price only one of them, and the two that hurt most are the two that never appear on a fee schedule.

The layers, in the order you pay them:

1. The spread. What you give up between bid and ask on the venue where you buy. Zero on a tokenised money-market fund bought at primary; on a tokenised equity with a median day under $1M it is the largest cost in the stack, and invisible on the screen.
2. The venue fee. A marketplace commission on a collectible token, or an exchange taker fee, typically in the 0.1% to 2.5% range depending on where you are. Published examples at the low end: Collector Crypt takes 2% of the seller’s proceeds and Courtyard takes nothing on an ordinary listing.
3. The network fee. Cents on Solana or Polygon, dollars on Ethereum mainnet at ordinary 2026 congestion. Trivial for a $10,000 position, punitive if you are rebalancing a hundred $86 cards.
4. The wrapper’s own fee. A fund management fee, a vault fee, a certificate issuer’s spread. This is the one that compounds, and it is the one to interrogate hardest.
5. The exit. Redemption charges, delivery and shipping on a physical, or the discount you accept because the secondary market is three people.

Tax follows what the token looks through to, not the fact that it is a token, and the rate that surprises people is the one on gold.

A tokenised money-market fund pays interest, which is ordinary income, reportable annually as it accrues whether or not you sell. A tokenised equity wrapper is in most structures a debt or derivative claim rather than a share, so its treatment can differ from owning the stock; that is a question for a professional, and remember the 30% withholding already taken out of the reinvested dividend before you get there.

A tokenised collectible runs into IRS Notice 2023-27 of March 21, 2023, which applies a look-through analysis: if the underlying is a collectible under section 408(m), the token is treated as one, and long-term gains are taxed at a maximum of 28% under section 1(h)(4)–(5) rather than 20%, plus the 3.8% net investment income tax where it applies. No final guidance has been issued as of September 2026. A graded card is squarely a collectible. A tokenised Treasury fund plainly is not.

The case most readers miss is gold. Section 408(m) lists “any metal or gem” alongside art, rugs and stamps, so tokenised gold is a collectible too, taxed at up to 28% rather than 20%. Physical bullion and the bullion ETFs pay the same rate, so this is not a penalty the wrapper adds; what it does is take 8 points out of any pitch that puts metal in a token and calls the result an ordinary capital asset. And a collectible cannot sit in an IRA without a deemed distribution, which rules both gold coins out of the account most readers would want to hold them in.

Three mechanical points close the section. Buying a token with appreciated ETH or SOL is a taxable disposition of the crypto, at gain, before the token has done anything; the Investing in NFTs guide works that trap through in detail and it applies identically here.

Form 1099-DA now requires custodial brokers to report gross proceeds on sales from January 1, 2025, and cost basis on digital assets acquired on or after January 1, 2026 and held continuously in the broker’s account, so the reporting gap of 2021 has closed for anything you buy from here.

And section 1091, the wash-sale rule, still reaches only “stock or securities” and has not been extended to digital assets as of 2026, so a loss on a tokenised commodity or collectible can be realised and the position re-established immediately. Two live proposals would close it — a Senate digital-asset tax bill from 2025 and a bipartisan House discussion draft from December 2025 — and neither is law. Assume it will not last.

Running $20,000 through both ends of the market for three years shows where the money goes, and on the platforms’ published fee schedules the answer is not the one an assumed fee stack gives: the wrappers cost very little, and almost the whole gap to a Treasury bill is the cost of holding an asset that pays nothing.

Leg A, the tokenised Treasury fund

You buy $10,000 of an accessible tokenised government money-market fund. Assume the bill yield holds at the 3.625% midpoint of the July 2026 target range, which was still the range on September 10, 2026; that the fund charges 25 basis points, in the middle of the 15-to-50 range we surveyed in May 2026; and that a mint and a redemption cost $5 each in network fees.

• Gross at 3.625% for three years: $10,000 becomes $11,127.40.
• At the net 3.375% you actually receive: $11,047.06. The fee has cost $80.34.
• Less $10 of network fees: $11,037.06. Profit in hand: $1,037.06. Network fees are not deductible against interest, so the tax base is the $1,047.06 the fund paid you.
• Interest is ordinary income. At a 35% federal marginal rate plus the 3.8% net investment income tax, 38.8% of that $1,047.06 is $406.26.
• Net after tax: $10,630.80, a 2.06% compound annual return.

Leg B, the tokenised cards

You buy at our tape’s September 8, 2026 Collector Crypt median asking price of $85.98, on that platform’s published schedule: no buyer-side platform fee, a 2% fee taken out of the seller’s proceeds, a 2% vault-withdrawal fee on redemption, nothing for storage or insurance, and Solana network fees of about a dollar in total.

• $9,974 buys 116 cards at $85.98, with about $1 of network fees.
• Hold three years. Assume the cards are flat, which is the honest base case: there is no long-run index for sub-$100 graded cards, and over our own eleven-day reading listings rose 9.4% against a median ask that ended 4.5% lower and swung between $75.00 and $115.00 in between.
• Sell at $85.98 each, less the 2% platform fee: $9,774.21. Loss: $200, or 2.0% of what you put in, entirely to fees — and that assumes you sell at the same ask you paid, which ignores the spread. Take the platform’s standing buyback at 85% to 90% of indexed value instead and the loss is 10% to 15%.
• The alternative exit is redemption. The 2% withdrawal fee on the whole position is $199, plus shipping and handling, which neither platform publishes as a flat rate. Redeem all 116 in one insured consignment and it is cheap; redeem them one at a time and shipping alone can exceed the value of an $86 card.
• The loss is a capital loss, deductible against capital gains and up to $3,000 a year of ordinary income.

Together: $19,975 in, $20,405 out after three years, a gain of 2.2% before inflation. The same $20,000 in three-month bills at 3.625%, taxed at 38.8%, would have produced $21,380. The gap is about $975.

Decompose that gap and the wrapper is barely in it. Half of the bill leg, $10,000, nets $10,690 after tax over the three years. The tokenised fund leg finished at $10,631, only $59 behind it: 25 basis points of management fee and $10 of network fees, most of it handed straight back by the tax you did not pay on the yield you did not earn. The other $916 is the card leg, which is $10,000 sitting flat for three years while the other $10,000 compounded.

That is the finding, and it is not the one assumed fees produce. On the published schedules the card leg loses 2.0% to fees rather than the 6.9% a 2% buyer’s fee and a 5% seller’s fee would have cost, and a full redemption costs 2% of the position rather than the 29% a $25-a-card charge would have implied. The collectible wrapper is far cheaper than its reputation, and the invented fee stack was doing most of the argument’s work. The bill still wins anyway, and it wins on yield rather than on fees: a cheap wrapper around a flat asset is still a flat asset.

Four failures have actually destroyed capital in this market, and none of them is the one the sceptics usually name.

The wrapper fails, not the asset

Your gold is fine; your issuer is not. Read where the underlying is held and in whose name: allocated, segregated, held in trust for token holders, with a named custodian and published bar lists, is a genuinely different position from an issuer’s promise backed by an unspecified pool. RealT is the cleanest instance, set out in the section on the illiquid end above: a tokenised-real-estate platform in voluntary liquidation since July 2, 2026, with reported estimates of about $640,000 in investor escrow against roughly $140M raised. Every token transferred correctly the whole way down. The houses, the compliance and the company did not.

Redemption is gated exactly when you need it

Every fund and every vault reserves the right to suspend redemptions, and every one of them will use it in a crisis, which is the moment it matters. The question to ask before you buy is not whether the right exists but whether anyone has exercised it at size and what happened.

The price feed is wrong

Where a token is used as collateral, a stale or hard-coded oracle can keep a lending market open long after the asset has moved. In November 2025 the yield-bearing token xUSD went on being priced at a hard-coded $1.00 in vaults on Morpho, Euler and Silo while it fell 77% in a day and then to seven to fourteen cents, after its manager disclosed a $93M loss; the depositors who lost money had never heard of the underlying strategy, and the contagion across connected protocols was put at $285M and up. Read that as a general warning about any tokenised asset whose on-chain price is asserted rather than traded.

Somebody takes the keys

The theft record is the ordinary crypto one and it is large. Chainalysis counted $3.4B stolen across crypto in 2025, of which one incident, the February 2025 Bybit hack, was $1.5B. The first half of 2026 ran to roughly $1B of exploits on the counters that publish one — Immunefi put it at $972M across 207 incidents, a record count, Blockaid above $1B, CertiK at $1.3B — and compromised private keys overtook smart-contract bugs as the leading vector for the first time on record.

The pattern case is Resolv, on March 22, 2026: an attacker who had compromised the firm’s AWS key-management environment obtained the signing key that controlled the minting contract, deposited about $200,000, and minted roughly 80 million unbacked tokens because the contract validated neither the price ratio nor a maximum, cashing out around $25M. Phishing is the retail version: ScamSniffer counted $83.85M drained from about 106,000 victims in 2025, down 83% from $494M in 2024. A tokenised asset is a bearer instrument to a thief even when it is a registered security to a lawyer.

What is not on this list is the risk people expect: the chain going down, or the token being technically invalid. The chains have worked. Every loss in this market has been a human, legal or custodial failure wearing a technical costume.

There is a sequence that keeps you out of the four failures above, and it is deliberately slow.

1. Decide whether you want the asset at all. Read this house’s guide to the underlying first: gold, sports cards, private credit, real estate. If you would not buy the asset in its ordinary form, the token is not a reason to.
2. Identify which of the three models the product uses. Register, warehouse receipt or wrapper. The offering document says; the marketing page does not.
3. Write the redemption sentence. Who is obliged, on what notice, at what minimum, for what total cost. If you cannot, stop.
4. Price the full stack. Spread, venue fee, network fee, wrapper fee, exit cost. Compare that number with the ordinary version of the same exposure: an ETF, a brokerage account, a dealer.
5. Check eligibility before you fund anything. Most tokenised securities are restricted to whitelisted wallets and accredited or non-US investors. Find out whether you can hold it before you own the crypto you intend to buy it with.
6. Buy the smallest size the venue allows, and redeem it. Run the round trip once, at a size you do not care about, and record what it cost and how long it took. This step tells you more than any amount of reading.
7. Custody deliberately. Hardware wallet for anything you hold yourself; a chartered custodian for size. Keep the acquisition lot record for the crypto you spent, because that disposition is taxable.
8. Transact in the underlying’s market hours. For anything with an off-chain benchmark, this is free money relative to trading at 4am.
9. Size it as the risk it is. A tokenised bill fund is cash with an operational risk. A tokenised card, building or loan is an illiquid alternative asset and belongs in whatever sleeve you have set aside for those, not in the cash sleeve because it settles quickly.

Five readings would change the view, and each has a threshold and a date you can check it against.

The category’s size

rwa.xyz’s tokenised real-world assets excluding stablecoins were about $34.7B in August 2026, having read $35.2B on July 10, 2026; other trackers published $29.5B to $37.9B in the same weeks, so watch the trend on one tracker rather than the level across several. A sustained move above $50B would mean the category has become something other than one institutional cash product; a fall through $25B would mean the 2025–2026 expansion was a rate cycle.

Whether yield beats convenience on-chain

Tokenised Treasury funds at $16.2B on August 3, 2026 were about a twentieth of the $301.7B of dollar stablecoins on September 3, 2026. If that ratio passes 15%, on-chain dollars have started to care about interest, and the tokenised fund becomes the default cash instrument rather than a corporate-treasury product.

Access

The single most consequential change for a retail reader would be a large tokenised money-market fund dropping its minimum below $100,000, or a registered 1940 Act fund launching an on-chain share class with retail access. Franklin Templeton’s BENJI is the existing counterexample, and on August 12, 2026 the SEC cleared Franklin Templeton to use the BENJI system for cash management inside its other funds — a sign the machinery is being trusted with more, though not itself a retail opening. Watch for a second retail share class.

The rule

Regulation Crypto Assets was proposed on August 18, 2026. Watch whether it is adopted, and whether the final text keeps the safe harbour and the $75M-a-year exemption. Watch the CLARITY Act’s Senate cloture vote of September 15, 2026 and any reconciliation that follows.

Our own tape

Two readings we publish are direct measures of this market’s health. Tokenised gold outstanding was 1,178,048 ounces on May 23, 2026 on our Coin Metrics computation; ounces falling while the gold price rises means holders are redeeming, which is the sign of a working warehouse receipt rather than a failing one. And Collector Crypt’s listing count against its median asking price, 142,884 listings at $85.98 on September 8, 2026, is a supply-and-demand read on tokenised collectibles: listings rising faster than the median ask over a quarter would say the vaults are filling faster than buyers are arriving.

Every wrapper above has been described by what it costs and what it has done rather than by what it promises. The rule that survives all of it: a tokenised asset carries the risk of the underlying plus the risk of the wrapper, and only one of those two is on the chart.