Unlevered, falling like leverage: five neocloud and datacenter tokens

TL;DR

The five instruments closest to the AI buildout rank among the worst on the book over 30 days — and not one of them carries leverage.

• 30-day declines run −20.90% to −42.42% against a book median of −16.71%. Four sit inside the worst 15 on the book: IRENB is 8th from the bottom, CBRSB 9th, CRWVB 12th, NBISB 15th.
• Only seven instruments fell further than IRENB, and five of them are leveraged ETFs (INTWB −78.03%, SNXXB −61.18%, SOXLB −51.90%, MVLLB −51.66%, KORUB −49.66%). These five are plain common stock — no daily reset, no multiplier.
• They fell the most and were traded the least: group turnover 0.4927x against 0.6659x for the other 95. They hold 1.9% of market cap and do only 1.4% of volume.
• The counter-example is SMCIB. Listed the same day as IRENB with the identical 88-day window, it is −20.90% over 30 days against −42.42%, and 1.557x wide against 2.133x. The widest name in that cohort is a rare-earth miner, USARB at 2.313x, which is not on this layer at all.
• The data has two holes in this group. The listing status of CBRSB could not be established at ingest, so its type reads unclassified and its underlying symbol is empty — and it carries a $5.42M market cap and a −40.09% 30-day decline. The $12.82M market cap on NBISB comes from price × circulating supply; the upstream payload reports a total supply of zero, so read as a fully diluted figure this instrument becomes $0.
• This snapshot cannot produce a premium. It carries no underlying reference price and no dividend-reinvestment multiplier, so this piece publishes no premium or discount figure at all — not an omission, an impossibility.

1. What these five are

bStocks is Binance's in-house tokenized-equity issuer (BTech Holdings Limited, ADGM / FSRA listed certificates, BNB Chain only, traded on Binance only). CoinMarketCap reports 105 instruments; the snapshot behind this piece holds 100 of them, timestamped 2026-10-31T06:06:52Z.

These underlyings are not companies that benefit from AI. They are the physical implementation of the compute buildout: a neocloud renting GPUs, an inference silicon designer, an operator converting mining sites into datacenters, a server manufacturer.

Together they hold $25.35M of the book's $1,318.38M — 1.9% — and $12.49M of its $873.46M of volume, 1.4%. Turnover is 0.4927x against 0.6659x for the other 95 — the rest of the book turns over 1.35 times faster than this group.

One shape inside the group is worth recording first: NBISB is the largest by market cap ($12.82M, 50.6% of the group) and the lowest by turnover (0.375x), while SMCIB is the smallest ($1.36M) and the most resilient over 30 days. Size and performance run opposite here.

2. The worst tier on the book is almost all leveraged products — except this group

Of the 100 instruments, 82 are negative over 30 days with a median of −16.71%. Where this group lands:

Four sit inside the worst 15 on the book.

List the seven instruments that fell further than IRENB and the weight of that shows up:

Five of the seven are leveraged ETFs. A leveraged product falling that far is mechanical — a daily-reset 2x or 3x exposure amplifies a one-way decline by construction. The five here are common stock: no multiplier, no daily reset, no decay, and the same range of outcomes.

The boundary of that claim needs stating: this does not say the risk of these five equals that of a leveraged product. It says that in the window from 2026-10-01 to 10-31, an unlevered sector produced declines that normally require leverage. The decline comes from the underlying stocks, not from the structure of the token.

The in-range figures for the other four run from 2.7% to 9.5% — all four are pinned to the bottom of their own range since listing. Only SMCIB is at 22.9%.

3. The counter-example: SMCIB, and the wider names are not on this layer

The two tables above invite the reading that the datacenter layer is simply the worst and most volatile one. First, the ranges (high ÷ low since listing) for all five:

"Rank of 46" is the position from narrowest to widest among the 46 instruments listed 100 days or more; IRENB and SMCIB have been listed 88 days and are not among those 46.

Two counter-examples.

The first is SMCIB. Listed the same day as IRENB (2026-08-04), the same 88-day window, the same hardware side of the buildout underneath:

• −20.90% over 30 days against −42.42% for IRENB — 21.5 percentage points apart.
• Range width 1.557x against 2.133x.
• In range 22.9% against 4.2%.
• +0.12% over 24 hours and +0.16% over 7 days — the only member of this group that did not fall in either window.

Same cohort, same layer, same window, and the outcome differs by a factor of two. "AI buildout stocks got taken down this month" holds for four of them and fails on the fifth.

The second is the neighbourhood in the width ranking. NBISB and CBRSB both listed on 2026-07-06, one of ten instruments that day:

The 1.996x on NBISB sits below a hard-drive maker, WDCB at 2.005x, 0.5% apart. An AI neocloud and a legacy storage hardware name covered the same amount of ground over the identical 117-day window.

And the widest member here, CBRSB at 2.530x, is still narrower than SOXLB at 2.892x from the same cohort — a 3x leveraged semiconductor index ETF. So "this layer is the most volatile" only comes close to holding once leveraged products are excluded, and section 2 already showed how large a concession that exclusion is.

4. The data has two holes in this group

This snapshot happens to be unclean in two places in this group. Neither affects the conclusions in sections 2 and 3, but both affect the premise of what these numbers are.

The first hole: the type on CBRSB reads unclassified. Eight of the 100 instruments carry that type and CBRSB is one of them, the only one in this group. The meaning is specific: the listing status of the company could not be established at ingest, so the field is neither common stock nor private but left blank. Its underlying ticker symbol is empty for the same reason — the ingest rule does not derive a symbol by stripping the trailing B from the token name, because that rule is right for most rows and wrong for BRK.B, for Korea-listed SK Hynix and for private SpaceX.

So the state of the CBRSB row is this: an instrument with a $5.42M market cap, a −40.09% 30-day decline and the widest range in this group at 2.530x, and the data does not claim to know which market its underlying trades on or in what form.

The second hole: the market cap on NBISB depends on which field you read. Its $12.82M is price $155.97 × 82,191 tokens in circulation. The upstream payload reports a total supply of zero, so the fully diluted market cap computed from that supply is $0 — read the wrong field and the largest instrument in this group, 50.6% of its market cap, disappears from the statistics entirely.

That one needs its source labelled: it was recorded at ingest and cannot be recomputed from the committed snapshot itself. The snapshot stores only the market cap, not the total supply, so what this piece can confirm is that the market cap uses the circulating-supply definition; it cannot reproduce that zero from this file. It is worth raising because it demonstrates something more general: the same instrument can differ by everything across two fields.

The third hole is a matter of definition and applies to all five: the snapshot stores the high and low since listing, not the all-time high and low. The windows here are 88 to 117 days, so "−58.8% from high" means from the highest price since listing, not from an all-time high. What the underlying stock did before the token listed is not in this data.

5. Putting the group back in the book

Cut by listing month:

Three of this group listed in July (NBISB / CBRSB / CRWVB) and two in August (IRENB / SMCIB).

On size: across the book only 7 instruments are above $50M and 12 above $25M, while 61 sit below $5M and 33 below $1M, with a median of $2.39M. The largest here, NBISB at $12.82M, ranks 19th; the smallest, SMCIB at $1.36M, ranks 61st. The whole group lives in the middle.

One caliber trap: turnover = 24h volume ÷ market cap, and a small denominator inflates the ratio mechanically. This group turning over below the book says first that their market caps are large relative to their volume that day, and cannot be read directly as nobody wanting them — though the declines in section 2 make the same point from the other direction.

6. Structure: all five share one wrapper

• The wrapper is a tracker certificate, not equity. The issuing entity is BTech Holdings Limited, ADGM / FSRA listed certificates.
• No voting rights.
• Dividends are passed through: a 30% US withholding deduction first, then reinvestment through a multiplier (see section 8).
• Reserves are covered by attestation, not an on-chain proof of reserve. Among comparable issuers xStocks uses Chainlink proof of reserve, which updates daily or on a 10% move in reserves and can be checked on-chain. An attestation cannot.
• Redemption requires KYC, with a $5 minimum.
• US persons are blocked.
• One chain. BNB Chain only, BSC deposits only, so a wrong-network transfer loses the asset.

This is a description of structure, not an allegation. But placed next to section 4 it points at something concrete: the reserves behind CBRSB are backed by an attestation, while this data has not established which market its underlying trades on. Those two facts do not contradict each other — the attestation is the issuer's, the type field is the ingest step — but for a reader, verifiability is thinnest on that row.

7. The snapshot was taken on a Saturday

The timestamp 2026-10-31T06:06:52Z is a Saturday morning UTC. The US cash market closed Friday at 20:00 UTC and does not open again until Monday.

The same issuer has already been measured on this point once: one stock, four books laid the four tokenized versions of AAPL side by side on a Saturday and concluded that a quote outside trading hours is a market maker opinion, not a market price.

The 24-hour window here (Friday 06:06 UTC to Saturday 06:06 UTC) contains one full US session, so it is not a pure closed-market quote. But the snapshot itself lands inside the close, and for the two-plus days from there until Monday, these token prices are the only prices these names have.

One extra layer for this piece: the listing status of the CBRSB underlying is not established in this data, so whether "the US market is closed" even applies to it is itself uncertain. The other four underlyings are US-listed and the condition holds for them unambiguously.

8. What this method cannot calculate

First, no premium or discount. The snapshot carries only the token price and no reference price for the underlying. More fundamentally, the bStocks dividend mechanism takes a 30% US withholding deduction and then reinvests through a multiplier — without knowing the current multiplier, the token unit price and the underlying share price are not comparable quantities. So this piece publishes no premium figure. That is an impossibility, not an oversight.

Second, no shares per token. The snapshot has no conversion ratio. The 82,191 NBISB tokens in circulation cannot be converted into any number of shares without it.

Third, the range-width windows are not equal length. This group runs 88 / 88 / 103 / 117 / 117 days. Book-wide comparisons use only the 46 instruments listed 100 days or more — IRENB and SMCIB are not among those 46, which is why section 3 compares their widths only against instruments listed the same day. Even inside those 46 the windows differ by over a month.

Fourth, market cap is the CMC definition: price × circulating supply. The book total of $1,318.38M here is these 100 rows added up, while the /stocks page on this site shows $914M of on-chain value for the bStocks row — two figures from different sources at different moments, and they will not reconcile. This piece uses the former throughout and timestamps it everywhere. The zero total supply in section 4 is the extreme version of the same thing: switch the definition and the figure can go from $12.82M to $0.

Fifth, 100 of 105. CMC reports 105 instruments, the snapshot holds 100, and the missing 5 are in none of these statistics.

Sixth, no attribution of the decline. Section 2 can show that these five land in the same range of declines as leveraged products. It cannot decompose that decline into the underlying stock falling, selling pressure on the token side, or market makers pulling quotes. That needs an underlying reference price and a tick-level book, neither of which is in the snapshot.

Seventh, "neoclouds and datacenters" is a layer this piece drew. The snapshot has no industry field, and CBRSB has no type field at all. Grouping these five rests on what each company does, not on any classification from the issuer or from CMC.

Conclusion. The underlyings here are the physical implementation of the AI compute buildout, and in this snapshot their 30-day declines run −20.90% to −42.42%, with four inside the worst 15 on the book. Only seven instruments fell further than IRENB and five of those are leveraged ETFs — an unlevered sector producing declines that normally require a multiplier. They are also colder than the book: 1.9% of market cap, 1.4% of volume, turnover 0.4927x against 0.6659x.

But this is not a rule about datacenters. SMCIB and IRENB listed the same day for the same 88 days and are 21.5 percentage points apart over 30 days; the range width of NBISB sits below a hard-drive maker by 0.5%. What decides these numbers is the specific company, not the label they share.

And one thing more basic: the data has holes in this group. The listing status of CBRSB could not be established at ingest, and the market cap of NBISB depends on which field is read — the first means a $5.42M instrument whose underlying trading form this data does not claim to know; the second means the same instrument can differ by everything across two fields. Before ranking anything by these numbers, establish which definition they are on.

Data from CoinMarketCap, timestamped 2026-10-31T06:06:52Z, covering 100 of the 105 instruments bStocks reports. Instrument detail and quotes at /stocks/bstocks. Research notes, not investment advice. DYOR.