Tokenized stocks

The same stock is tokenized by many companies at once. Eight issuers offer a tokenized NVIDIA. They are separate products with separate custodians and separate redemption mechanics, trading in separate markets.

Which one is safe to hold is the question a holder actually has, and it is not answered by a list of products. It needs a per-issuer read on where each version trades and how far its price sits from its peers. That is what this page is.

Four markets, one company, no shared oracle

Correction, 2026-07-28. This section used to be headed "Price where none exists" and opened "SpaceX is private. There is no NYSE quote." SpaceX listed on Nasdaq as SPCX on 2026-06-12, and that claim was wrong here for six weeks. It is corrected rather than deleted, because the page it sits on is about not publishing confident wrong answers.

What is still true, and still measured, is the agreement itself:

Four issuers, four independent order books, no shared oracle between them, pricing the same company within cents of each other. That is a real finding about how tokenized equity markets behave, and it does not depend on the underlying being private.

Answered, 2026-07-29. The equity feed is built, so this page no longer stops at comparing issuers to each other. raw.underlying_prices carries the listed daily close and every row below can be measured against the real share.

And the open question about SpaceX resolves the dull way. This section used to note that these markets priced SpaceX near $110 while the IPO priced at $135, and refused to call it a prediction error without establishing the share-unit basis first. That caution was right, and the answer is that there was never a discrepancy: SPCX has since traded down to $116.41, and the tokenized version quotes $114.85, within 1.3%. The $110-versus-$135 gap was a current price compared against a historical one, which is the same class of mistake as comparing a single-chain supply to a multi-chain total. The units were fine all along.

Premium against the real share: withdrawn

This page used to publish a premium for every tokenized stock against the real listed share. It has been removed because it was wrong, and the way it was wrong is worth stating rather than quietly deleting.

Two faults, both measured 2026-07-31.

The reference was stale and the guard did not catch it. Every row compared a live 24/7 token quote against a stock close from up to 37.8 hours earlier. The staleness ceiling was set at 96 hours, so all 183 rows passed as daily_close while the equity feed sat at one successful run against hourly token prices. Bloom Energy published +35.04%. Eleven boards read above +14% at once. None of it was a premium; it was a full trading session of ordinary movement charged to the tokens.

And some token quotes are simply wrong. After refreshing the reference to 13.8 hours the worst case barely moved: Ondo's tokenized CrowdStrike still reads a 300% premium, quoting about $741 against a real share near $179. That is not staleness, it is a bad price, and no reference age fixes it.

The honest version of this comparison needs an equity price sampled close to when the token was quoted. A daily close cannot do it, because the whole point of these instruments is that they trade while the exchange is shut. Blue Ocean ATS covers 20:00 to 04:00 ET and is the obvious input; the weekend remains genuinely uncovered by any US venue.

Until then there is no premium column, which is the same call already made once on the pre-open signal for the identical reason.

Does the 24/7 market know where the stock opens?

This is the strongest claim made for tokenized equities: the exchange shuts, the token keeps trading, so the token should tell you tomorrow's price. We measure it rather than repeat it — and the first version of this section got it wrong, so it is worth saying how.

We were scoring against the wrong price. A quote taken at 2am ET is predicting the opening bell, not 4pm. Measured against the close, the token looked useless and this page concluded it was "pinned near the prior close rather than discovering a price". Measured against the open, which is what it is actually forecasting, the median realized move is 1.01x what the token implied. It sizes overnight gaps nearly correctly. Against the close the same quotes give 2.00x, and that gap is not a failure of the token: it is the trading session the token could not have known about.

We still will not give you a hit rate. Against the open the direction is right on 93 of 98 name-days, which looks spectacular and is not evidence. On each day measured, 85-96% of the tokenized signals pointed the same way and 74-96% of the opens did too, because both are dominated by one market-wide overnight move. That is three independent observations, three of three, not 98 — and the same pooling gave a 63% figure at p=0.046 a day earlier that evaporated when a third day arrived. The magnitude numbers above need no such assumption, which is why they are the ones published. Ask again in a month, when the sample is days rather than a long weekend.

And one more limit, because it is the one a professional would raise first. US stocks already trade overnight on Blue Ocean ATS, 20:00 to 04:00 ET Sunday through Thursday, since 2021. Every snapshot behind the numbers above but one was taken inside those hours, so a conventional US venue was pricing the same share at the same moment and the token may be tracking it rather than discovering anything. Split by whether any US venue was open, the weekend observation sizes the gap at 1.03x and the Blue Ocean-overlapping ones at 1.25x — but that is one day against two. The hours genuinely nothing else covers are Friday 20:00 to Sunday 20:00 ET, and that is where this dataset can say something no other source can. We are measuring against Blue Ocean before claiming more.

Which version should you hold?

Issuers to avoid

Two independent failures at once: no meaningful volume and a price adrift from peers. An issuer can wash trade its own token, but it cannot make competitors agree with a wrong price, so failing both is a strong signal.

Qualified liquidity

Every liquidity figure on this page is a qualified figure: the sum of pools that pass a stated test, applied per pool, with the excluded amount published beside it so the derivation is checkable.

The test is a plausibility ceiling rather than a size floor. A pool claiming more than 10x the market capitalisation of the token it holds does not qualify, because a pool cannot hold ten times as much of a token as exists.

On Robinhood Chain all pools together come to $2.97 billion and the qualified figure is $33 million. One pool is the entire difference, and its own numbers are what the test reads:

• it claims 508.7x the market capitalisation of the token it holds, $2.93B of depth against a $5.77M token
• it has recorded zero trades in 24 hours
• it quotes NVDA at $2.87, against a real closing price near $197

The nvda in that pool's name is not NVIDIA. It is a token named to resemble it, which is why every comparison on this page matches pools by contract address and never by pool name: two different tokens on Robinhood Chain both present as NVIDIA.

Where the 10x came from

A threshold picked for looking round is not a standard, so this one was calibrated against the data and against its own failure mode.

Market capitalisation is not a hard bound in practice. CoinGecko derives it from circulating supply, which is routinely understated for small tokenized wrappers, so pools that are entirely honest still land at 1.5x to 3x. A ceiling set at 1x flagged three pools and one of them was real: QQQB on pancakeswap-v3-bsc, at 1.8x, quoting $678.61 against a $690.63 reference and turning over $86.7M in 24 hours. That is an actively traded market, and a 1x ceiling would have deleted $2.37M of genuine depth from the published figure.

So 10x sits well above the honest range and well below what the failing pool claims. The choice is deliberately reluctant in one direction: excluding real liquidity understates depth, and for anyone reading these figures as tradeable that is the more damaging error.

On Solana, Ethereum and BNB Chain nothing is excluded at all. That is the check on the ceiling itself. A test that fired on small venues for being small would show exclusions everywhere; one that fires only on a claim that cannot be true shows them where a claim cannot be true.

Where these actually trade

A token existing on a chain tells you nothing about whether it trades there. A dead pool is worse than uninteresting: its stale price still gets quoted. One such pool priced NVIDIA at $1,180 against a real $208, a 467% premium on zero volume.

They exist everywhere and trade almost nowhere.

Ghost listings

Deployed, quotable, and with no live venue at all:

How this is judged

A venue counts as live only if it moved at least $10,000 in 24 hours and holds at least $100,000 of liquidity. Both thresholds are configuration, recorded on every row, so any number here can be traced to the floor that produced it.

Raised 2026-07-28, from $100 and $1,000. Those were not a floor. $100 of daily volume is noise, and $1,000 of depth means a $500 order moves the price against you, so a venue could pass the old test while being unusable. The new pair is the honest reading of "a retail-size trade clears here". It cut the live count from 44 to 23, which makes the finding below stronger, not weaker: the old threshold was flattering the data.

Participant counts are used rather than trade counts. One pool showed 657 transactions from 8 addresses, which counting trades would have called a busy market.

See methodology for the full set of assumptions.