The risks in bitcoin treasury companies, with the arithmetic

Start here: how far these shares can fall

The single most important thing to understand before anything else is that the equity is far more volatile than the asset it holds, in both directions, and that the asymmetry is measurable rather than rhetorical.

The bitcoin figures below are unusually well evidenced because they come from Strategy Inc's own Form 10-Q for the quarter ended 30 June 2026 (filed 3 August 2026), which discloses quarterly high, low and closing bitcoin prices on Coinbase — a filed disclosure rather than a price aggregator.

How to read that. The peak-to-trough figures are the same episode viewed two ways: the asset lost a little over half its value, the equity lost more than three-quarters of it. The calendar-2025 row is the starker demonstration, because it isolates a year in which bitcoin barely moved: bitcoin fell single digits while the equity fell 47.53% on a total-return basis (Slickcharts; year-end closes of $289.62 for 2024 and $151.95 for 2025 per stockanalysis.com). That is a fall of roughly five to eight times the asset's, depending which bitcoin price source you use. Sources genuinely disagree on bitcoin's exact 2025 return — roughly −6% using the Coinbase year-end closes in the 10-Q, about −8.7% on one commercial series — so the ratio is a range, not a point.

MSTR's 52-week low of $81.81 is separately corroborated: The Block reported a close of $82.16 and $81.80 after hours on 26 June 2026, the same day the company's enterprise premium to net asset value broke below 1.0. Two independent records of the same trough.

The upside is the same mechanism. This is not a one-way critique. The leverage that produced a 77.6% drawdown also produced roughly +346% in 2023 and +359% in 2024. Anyone telling you the structure only amplifies losses is describing half of it. The point of this page is that the amplification runs both ways and that the downside half is arithmetically knowable in advance.

One figure to avoid. You will see an all-time maximum drawdown for MSTR quoted at −99.86%. That reflects the collapse of a completely different business — the dot-com era software company — and has nothing to do with the bitcoin strategy. Comparing it to bitcoin's −93.07% all-time drawdown is a category error.

What these companies actually are

A digital asset treasury company (DAT) is an ordinary listed company that has adopted holding a cryptoasset — usually bitcoin — as its principal corporate purpose. Strategy Inc, formerly MicroStrategy, is the largest; it held 840,447 BTC at 23 August 2026 per its weekly 8-K filed 24 August 2026. The mechanics of the structure, the vocabulary, and the sector as a whole are covered in our explainer on digital asset treasury companies; if the underlying asset class is unfamiliar, start with what digital assets actually are.

The promotional version of the model is a flywheel: the shares trade above the value of the bitcoin held, the company issues new shares at that premium, buys more bitcoin than the dilution costs existing holders, the bitcoin-per-share figure rises, and the premium is reinforced. Told that way it is genuinely accretive, and for several years it genuinely was.

The reverse gear is the same machine running backwards. When the shares trade below the value of the holdings, every share issued hands the new investor more claim on the bitcoin than they paid for it, and existing holders' claim per share falls. Any explanation of the flywheel that stops before this point is defective, because the flywheel does not have an off switch — the company still has preferred dividends and coupons to pay, and issuing shares is how it pays them. The four weeks documented in the next section are that machine running backwards, on the public record.

Four weeks of filings: more claimants, fewer coins

Dilution below net asset value is usually argued about in the abstract. Between 27 July and 23 August 2026 it was filed with the SEC, week by week, in four consecutive 8-Ks. This is the clearest documented illustration of the mechanism available anywhere, and it is worth reading the table slowly.

All figures above are from Strategy's weekly 8-K filings of 3, 10, 17 and 24 August 2026 (Items 7.01 and 8.01); the last of them is here, and the full run is on the company's EDGAR filing history. Weekly figures include shares sold but not yet settled, so week boundaries can overlap very slightly.

What the table says in one sentence. In four weeks the company created about 31.3 million new claims on the treasury — roughly 8.2% on top of the 384.23 million shares outstanding at 24 July 2026 — raised about $3.28bn, bought not one satoshi, and ended the period holding 3,328 fewer bitcoin than it started with.

The per-share arithmetic. Gross bitcoin per share went from 843,775 BTC over about 384.23m shares — roughly 219,600 satoshis a share — to 840,447 BTC over roughly 415.5m shares, about 202,300 satoshis. That is a fall of about 7.9% in four weeks. This is our own calculation from the filed share and coin counts, not a figure the company publishes.

Where the money went. The 8-Ks state the uses of proceeds, and they are the point. Week to 2 August: $250.0m to the USD Reserve, $28.9m to repurchase STRC preferred stock, $11.7m to cash. Week to 9 August: $650.0m to the USD Reserve. Week to 16 August: $52.4m to fund STRC dividends, $132.2m to repurchase STRC, $149.1m to the USD Reserve. Week to 23 August: $136.4m to repurchase STRC, $300.0m to the USD Reserve, the remainder to a newly created "USD Cash" pool. Common shareholders were diluted to service and support the securities that rank ahead of them.

And the coins were sold below cost. The 3,328 bitcoin left the balance sheet at $63,957 and $64,262 against an average purchase price of about $75,385–$75,419 — roughly 15% below cost.

Why this is the textbook case of issuing below 1.0x mNAV. The market's valuation multiple on net asset value — mNAV — decides whether issuance helps or harms the existing holder. Above 1.0, a share sold for more than the net assets behind it brings in more value than the claim it creates, and everyone's coins-per-share goes up. Below 1.0, the trade reverses: the buyer receives a claim worth more than the cash they hand over, and that difference comes out of the existing holders. Strategy's enterprise mNAV broke below 1.0 on 26 June 2026. The issuance did not stop. Vetle Lunde of K33 Research put the mechanism as plainly as anyone: below net asset value, equity issuance gives away more ownership through undervalued shares than the value received in return (The Block, 17 September 2025).

The headline metric can point the other way. "Bitcoin Per Share" and "BTC Yield" are the company's own non-GAAP marketing metrics, and they are gross of everything that ranks ahead of you. In the six months to 30 June 2026, gross bitcoin per share rose about 5.6% (672,500 BTC over 312.062m shares at 31 December 2025, to 846,000 over 371.603m at 30 June 2026) and the company reported "BTC Yield" of 8.1% for the half (Form 10-Q, quarter ended 30 June 2026); by 26 July 2026 the year-to-date figure had itself fallen to 4.5%. Over exactly the same six months, on our own calculation from the two filed balance sheets, bitcoin per share after senior claims fell about 16.6% — from roughly 164,530 satoshis to roughly 137,260. Same company, same period, opposite signs. BitMEX Research reached the same conclusion independently on 22 July 2026, finding that a gross figure of 222,000 satoshis overstated common-equity exposure by about 39% against a claims-adjusted 135,700.

The company says so itself. Strategy's own disclosure in the 30 July 2026 8-K exhibit 99.1 states that these KPIs do not take account of existing and future liabilities including debt, nor of the preferential rights of preferred stockholders, that holders of those instruments have claims on the company's bitcoin senior to common holders, and that the metrics may therefore overstate bitcoin per share after senior claims. It adds that "BTC Yield" is not a yield in the traditional financial sense and that the KPIs should be used only by sophisticated investors who understand their limitations. When a company warns you in a filing that its own headline number may overstate your position, the honest thing to say is that the marketing metric is misleading as generally used — not because it is fabricated, but because it measures something other than what a shareholder owns.

None of that is fraud, and it is not being alleged here: the caveats are disclosed, in writing, by the company. But calling a growth rate a "yield" borrows the vocabulary of income products for something that pays nothing, and that is the kind of language worth testing before you act on it. Our investment scam checklist covers the general technique for interrogating a promotional claim — who is measuring, against what, over which window, and what has been excluded.

What actually happens to the common shares when bitcoin falls?

This is the part worth doing on paper, because it is not an opinion. Every input below is a filed figure from the consolidated balance sheet in Strategy's 10-Q for the quarter ended 30 June 2026. The arithmetic that follows is ours.

There were 371.603 million common shares outstanding at that date (351.963m class A plus 19.640m class B), so the residual works out at $80.59 a share. The look-through leverage factor is $49,672.1m of bitcoin supporting $29,948.0m of common equity — 1.66x.

The useful way to hold this in your head is that the senior claims are a fixed number sitting in front of you. Debt plus preferred, less cash, is $19,724.1m. The common shareholders own bitcoin minus $19,724.1m — and a subtraction of a fixed amount from a falling number falls faster in percentage terms, without limit, until it reaches zero.

So: a 30% fall in bitcoin implies roughly a 50% fall in the value attributable to common shareholders, and a 60.3% fall extinguishes it entirely. The equivalent bitcoin prices in the left column are scaled from the $58,714 close on 30 June 2026 used in the filing.

What this calculation is, and is not. It is a single-date liquidation identity: what the common would be worth if everything were realised at those values on that day. It is emphatically not a price prediction, not a forecast, and not a path model. Three caveats matter:

• The leverage factor itself moves. On the 31 December 2025 balance sheet the same calculation gives $58,854.0m of bitcoin, plus $2,301.5m cash, less $8,190.2m debt and $8,032.3m preferred — a residual of $44,933.0m and a leverage factor of 1.31x. Six months later it was 1.66x, because the preferred liquidation preference nearly doubled from $8.03bn to $15.46bn. Leverage in this structure ratchets up as more preferred is issued, and it ratchets up again as the asset falls, because the senior claims do not shrink with it.
• It does not compound cleanly over a period. You cannot take 54.1% and multiply it by 1.66 to get the equity's realised drawdown; the actual figure was 77.6%, because the leverage factor, the share count and the market's multiple were all moving at once.
• Zero equity value is not the same as bankruptcy. The residual reaching zero is an accounting statement about who owns what. What actually happens at low asset prices is governed by the cash obligations covered further down, not by this table.

How this differs from a leveraged fund. A daily-reset leveraged ETF rebalances its exposure every day, which produces path-dependent decay in choppy markets — the mechanism we set out in the warning on leveraged and inverse ETFs. A treasury company does the opposite: nothing resets. Its leverage floats, rising automatically as the asset falls, so exposure is at its highest exactly when the asset is weakest. Neither structure is "safer" — they fail differently, and the failure modes are worth knowing apart.

Why does the share price fall further than the arithmetic says?

Because the table above prices the net assets, and the market prices the share. The gap between them is mNAV — the multiple the market puts on net asset value — and it is a second, independent multiplier sitting on top of the leverage. Our guide to mNAV covers how it is defined and why basic and enterprise versions give materially different numbers; the short version is that a figure quoted without its definition is not worth much.

The two multipliers, illustrated. Suppose bitcoin falls 30%. From the table, the residual per share falls to about 50% of its previous level. Now suppose the market's multiple on that residual also falls, from 1.5x to 0.8x — which is a factor of about 0.53. The share price ends up at 0.50 × 0.53, or about 27% of where it started: a fall of roughly 73%, from a 30% move in the asset. Those multiples are illustrative round numbers chosen to show the arithmetic, not a claim about any particular company at any particular date.

The compression is documented, not theoretical. Strategy's enterprise mNAV — the version that nets debt, cash and preferred — dipped below 1.0 on 26 June 2026, meaning the market valued the largest treasury company in the world at less than the bitcoin it held (The Block, 26 June 2026). Some sources described the discount reaching 0.72x in late June and others only a brief dip below 1.0; because definitions vary so much, treat any specific value as contested unless it is paired with the definition used.

It is a sector-wide condition, not one company's problem. As at 2 January 2026, around 37 of the 100 largest treasury companies traded below net asset value, against roughly 25% in September 2025 (DL News, citing BitcoinTreasuries.net). Named discounts on that date: Strategy 17%, Twenty One Capital 17%, H100 Group 32%, Vanadi Coffee 61%, OranjeBTC at parity. K33 Research had put the sector mean mNAV at 2.8 on 17 September 2025, down from 3.76 in April 2025.

• The steepest documented collapse is NAKA (KindlyMD/Nakamoto Holdings), whose mNAV fell from 75x to 0.7x, with market value down 96% from peak. It then sold 600 BTC for roughly $48m to repay $45m owed to a lender, and authorised a $25m buyback of its own shares — the whole failure sequence in miniature: premium collapse, loss of equity funding, asset sales to service debt, buybacks to defend the discount, a shrinking treasury.
• A UK-listed comparator. The Smarter Web Company held 2,712 BTC after buying 11.89 BTC on 3 August 2026 at an average of £47,052, and traded at an mNAV of about 0.72 — a 28% discount — in late August 2026 per bitcointreasuries.net. Weeks earlier it had used nearly 178 BTC to settle a convertible obligation: a UK example of selling the asset to retire debt. Note that at least one widely used aggregator showed 7,500 BTC for the same company on 25 August 2026, which contradicts its own announcements. Aggregator data in this sector is unreliable; cross-check against the company's filings.
• Metaplanet (TSE:3350) closed at ¥305 on 24 August 2026, roughly 83% below its ¥1,895 all-time-high close of 16 June 2025, with a tracker showing 0.69x basic mNAV on 25 August 2026. It had already fallen below 1.0x — to about 0.88x — on 17 October 2025, which prompted a buyback programme. Its reported holdings differ materially between sources, so we are not quoting a coin count.

And nothing forces the discount to close. This is the structural point people miss. An exchange-traded fund has a creation and redemption mechanism that arbitrages the price back to net asset value. A treasury company has none. The only levers are buybacks — which require selling the asset, shrinking the treasury — or liquidation, which management controls and is under no obligation to pursue. A discount can persist for as long as the market wants it to.

What has to be paid in cash — and what happens when it cannot be

Bitcoin does not generate cash. Preferred dividends and coupons are paid in cash. That mismatch is the engine of everything on this page.

The preferred dividend burden

Strategy paid $400.661m of preferred dividends in the second quarter of 2026 alone, against $49.110m in the same quarter of 2025 — an 8.2-fold increase in a year. The half-year figure was $630.188m. The preferred liquidation preference grew from $8,032.3m at 31 December 2025 to $15,462.1m at 30 June 2026, with preferred shares outstanding rising from 78,183 thousand to 153,529 thousand.

Set that against the operating business. Total revenues for the second quarter of 2026 were $122.4m and gross profit $81.6m. The software business covers roughly one-fifth of a single quarter's preferred dividend. The dividends are therefore funded by issuing securities or selling bitcoin — not by earnings. That is a structural fact, not a criticism of the software business.

Five preferred series were outstanding: STRF at 10.00%, STRK at 8.00%, STRD at 10.00%, STRE at 10.00% (euro-denominated, listed in Luxembourg), and STRC on a variable rate that escalated through 2026 — 11.00% in January, 11.25% in February, 11.50% from March, and 12.00% for record dates on or after 1 July 2026, with payment frequency moving from monthly to semi-monthly.

Cumulative or not — check, do not assume. Whether a preferred series is cumulative is one of the most consequential terms in the whole structure. If it is cumulative, a skipped dividend does not disappear: it accrues as arrears that must be cleared before common shareholders can receive anything, so the obligation builds up in the background during exactly the periods when the company can least afford it. If it is non-cumulative, a skipped payment is simply gone — cheaper for the company, but skipping is a loud distress signal and typically triggers other consequences set out in the terms, such as board or voting rights. Strategy's five series do not all carry identical terms, and we have not restated the cumulative status of each series from a primary document here. The governing terms are in each series' certificate of designations, filed on EDGAR. Read the series you are actually looking at.

When a "stable" preferred stops holding its price

STRC was designed to trade around $100. It hit a record low of $71.40 on 26 June 2026, about 29% below par. The company's response was to raise the dividend rate to 12.00% and to buy the stock back: 288,930 shares between 20 and 26 July 2026 at an average of $86.53, a 13.47% discount to par, then a further 1,431,212 shares for $136.4m in the week to 23 August 2026. It stated an objective of STRC trading over time at $99 to $100 and said it would hold the 12.00% rate until the stock traded sustainably near par. By 24 August 2026 STRC was at $97.21. STRK, meanwhile, traded at $67.05 against a $100 stated amount — a 33% discount, giving an effective yield of about 11.93% on its $8 dividend.

The mechanism to notice: an instrument that needs an escalating dividend and continuous buybacks to hold its peg is transmitting stress, not absorbing it — and both the higher dividend and the buybacks are funded by issuing common stock and selling bitcoin.

The convertibles and their put dates

Convertible notes are often described as equity-like, because if the shares rise far enough the notes convert and never have to be repaid in cash. That description holds only while the shares are above the conversion price. At 25 August 2026, MSTR closed at $126.84 — below every one of the six initial conversion prices: 2030A $149.77, 2028 $183.19, 2032 $204.33, 2031 $232.72, 2030B $433.43 and 2029 $672.40. All six series were out of the money, which turns $6.71bn of notionally equity-like debt back into ordinary cash debt with hard dates. Conversion prices are the "initial" figures and are subject to anti-dilution adjustment, so current effective prices may differ slightly.

The hard dates are the holder put dates — the noncontingent right of a holder to require the company to repurchase for cash at 100% of principal:

The cash coupon on the notes is small — roughly $35m a year on our calculation from the stated rates. The coupons are not the problem; the put dates and the preferred dividends are.

Forced monetisation, and the promise that broke

The "never sell" premise no longer holds. Strategy's board authorised a BTC Monetization Program, and roughly $218.4m of bitcoin was sold during 2026 up to 26 July, explicitly to fund preferred dividends. The 10-Q's digital asset note records second-quarter sales of 1,395 BTC at an average of $59,663, removing $168.1m of original cost basis for $83.2m of carrying value. It states the purpose without ambiguity: "The proceeds from the sales were used to fund dividend payments on Preferred Stock."

The authorisation also covers funding the USD Reserve up to $1.25bn of sales, funding interest, and funding buybacks. At 23 August 2026 the company disclosed a USD Reserve of $5.10bn and a separate "USD Cash" pool of $1.59bn, the latter established explicitly as flexible capacity to repay, repurchase or redeem convertible notes. Building that buffer is a rational, defensive step — and it is also being built by selling shares below net asset value.

The company describes the failure mode itself. Its risk factors in the June 2026 10-Q state that if it depletes the USD Reserve and cannot obtain equity or debt financing in a timely manner, on favourable terms, or at all, it may be required to sell additional bitcoin to meet its obligations, potentially below its cost basis or on otherwise unfavourable terms; that such a sale could be read negatively by the market, pushing down the prices of its listed securities and impairing its ability to raise further finance; and that if it can neither raise financing nor sell enough bitcoin, it could be in default. That is the reflexive loop, in the company's own filed language, signed under SEC liability.

The 2022 precedents: these mechanisms are not hypothetical

Everything above describes how leverage on a volatile asset behaves under stress. The 2021–22 cycle is the evidence that it behaves that way in practice, and one of the clearest examples sits in Strategy's own filings.

The Silvergate collateral squeeze

On 23 March 2022, MacroStrategy LLC — a MicroStrategy subsidiary — entered a Credit and Security Agreement with Silvergate Bank for a $205.0m term loan at 30-day average SOFR plus 3.70% (floor 3.75%), maturing 23 March 2025, collateralised at closing by bitcoin worth about $820.0m. The agreement required a loan-to-value ratio of 50% or less, meaning at least $410.0m of bitcoin. If LTV exceeded 50%, the borrower had to deposit more bitcoin or prepay to bring LTV to 25% or less — or 35%, with a 25 basis point rate increase. On default, the lender could accelerate, add 2% interest, and liquidate the collateral.

Then the filings show what happened as bitcoin fell. Pledged collateral rose from 19,466 BTC at 31 March 2022 to 30,051 BTC at 30 June 2022 — a 54% increase in coins pledged, with no increase in principal. That is disclosed in the Form 10-Q for the quarter ended 30 June 2022.

Be precise about what this was. Michael Saylor publicly denied receiving a margin call in June 2022, and on the evidence available that denial appears accurate as stated. The filings do not show a margin call. What they show is a collateral requirement that forced a 54% top-up of pledged coins while the asset was falling. Not a margin call, but a documented collateral squeeze — and the distinction matters less than it sounds, because the cash-flow effect on the borrower is the same. Bitcoin subsequently fell to about $15,476 on 21 November 2022, below the roughly $21,000 level widely calculated as the original top-up trigger. The loan was repaid in full in March 2023 for $160m — a 22% discount to par — as Silvergate wound down.

The collateral chain of 2022

The individual failures are less instructive than the chain linking them:

• Three Arrows Capital was ordered into liquidation on 27–29 June 2022. It had defaulted on a Voyager Digital loan of $350m USDC plus 15,250 BTC. BlockFi and Genesis liquidated 3AC collateral after it failed to meet margin calls.
• Celsius froze withdrawals on 12 June 2022 and filed for Chapter 11 on 13 July 2022. It had lent 3AC $75m ($50m plus $25m).
• FTX ran the sequence in nine days: a 2 November 2022 report on Alameda's FTT-dependent balance sheet, Binance announcing an FTT sale on 6 November, roughly $5bn of withdrawals on 6–7 November, withdrawals halted on 8 November, bankruptcy on 11 November.

The mechanism worth teaching is the loop, not the names. Lenders liquidate collateral to protect themselves; the liquidation pushes prices lower; the lower price triggers further margin calls elsewhere; those trigger further liquidations. Leverage secured on a volatile asset fails precisely when the asset is falling — which is the same moment refinancing becomes unavailable. A treasury company is not margined in the way 3AC was, and that is a real structural advantage. But it does have dated cash obligations, and a dated obligation in a falling market is a soft version of the same problem.

The closed-end fund precedent

The Grayscale Bitcoin Trust traded at a discount of about 36.2% to net asset value on 30 September 2022, widening to roughly 49% by late December 2022, and spent most of 2023 around 40%. The instructive part is how it ended: the discount was closed by structural change, when conversion to an exchange-traded fund in January 2024 created a redemption mechanism that dragged the price back to net asset value. No such mechanism exists for a treasury company. The discount closes only if the market changes its mind, or if management sells assets.

One honest caveat. No bitcoin treasury company has yet entered insolvency — but three have already liquidated their treasuries in full. Satsuma Technology’s shareholders voted in July 2026 to sell all 668 remaining bitcoin, return the capital and cancel the listing; Bitdeer emptied its entire 943 BTC treasury by 20 February 2026; and Genius Group liquidated in April 2026. The other distress cases observed to date — NAKA, Sequans, ETHZilla, Semler Scientific — resolved through asset sales, buybacks or acquisition. The mechanisms described on this page are real and documented; the terminal outcome has not yet been observed, and it would be dishonest to imply otherwise.

Custody, counterparty, concentration and key-person risk

Everything so far has been about the capital structure. These risks sit alongside it and are independent of the bitcoin price.

Custody and counterparty

Strategy disclosed in its June 2026 10-Q that substantially all of its bitcoin sits with three US custodians: Anchorage Digital Bank N.A., Coinbase Custody Trust Company LLC, and Fidelity Digital Assets NA. It states that its primary counterparty exposure in respect of the bitcoin is to those custodians' performance of their obligations, that its custody contracts do not restrict reallocation between them, and that holdings may therefore be concentrated with a single custodian from time to time. The split between custodians is not disclosed, so the actual degree of single-custodian concentration cannot be assessed from outside. Its own risk factors point to the October 2021 Coinbase account-recovery exploit, which affected at least 6,000 customers, as an example of what custodial failure looks like.

There is a second-order point worth noticing: Coinbase is also the company's declared principal market for fair-value measurement under ASC 820. The same venue is simultaneously a custodian of the asset and the price source for the balance sheet carrying value.

Concentration that runs both ways

Strategy alone held 840,447 BTC at 23 August 2026 — about 4% of the 21 million supply cap. Public companies collectively held over 1.26 million BTC, more than 6% of supply, per bitcointreasuries.net aggregation. A holder of one of these shares therefore carries two distinct concentration exposures: to a single company's capital structure, and to the possibility that other treasury companies become forced sellers into the same market at the same time. NYDIG's Greg Cipolaro made the point directly in "How DATs Die" (15 August 2025): companies trading below net asset value face pressure to sell crypto to repurchase stock, which pressures crypto prices generally and accelerates the spiral.

Key-person and control

At 30 June 2026 there were 19,640,250 class B shares outstanding, each carrying ten votes against one vote for a class A share, representing about 35.8% of total voting power. Michael Saylor beneficially owned 19,616,680 of them. The company's own risk factor acknowledges that he has significant influence over matters requiring stockholder approval — mergers, going-private transactions, other extraordinary transactions and their terms, director elections, and amendments to the certificate of incorporation — and could impede a third party from acquiring the company.

Note the structural asymmetry this creates for a common shareholder: continuous class A issuance dilutes your economic ownership immediately and continuously, and dilutes the concentrated voting bloc only arithmetically as the class A count grows. And beyond the formal control, several independent analysts have argued that a significant part of the market premium these companies command is attached to a founder figure rather than to the balance sheet — which is a risk with no line item.

Regulatory and structural risk

Index inclusion and exclusion

Index membership is a source of price-insensitive buying. Losing it is a source of price-insensitive selling. In late 2025 MSCI consulted on excluding digital asset treasury companies from its equity indices, targeting companies where digital assets are 50% or more of total assets and the primary business is described as digital asset treasury. Reporting at the time put the scope at 39 companies with about $113bn of combined float-adjusted market capitalisation — 18 existing constituents removed and 21 blocked from future inclusion — with analyst estimates of $10–15bn of resulting outflows. Those outflow figures are estimates, not measurements.

On 6 January 2026 MSCI decided not to exclude them, and MSTR rose about 6% after hours. But the risk was deferred, not resolved. MSCI also deferred share increases, inclusion factor changes and size-segment movements for these companies in the interim, and signalled a broader consultation on the treatment of non-operating companies that resemble investment funds. The structural question — whether a company that mostly holds one asset belongs in an equity index at all — remains open. We were not able to retrieve MSCI's own published decision document; this account rests on secondary reporting and should be checked against MSCI's index consultation results before being relied on.

Accounting and tax treatment

ASU 2023-08 requires crypto assets to be measured at fair value with changes running through net income. The effect on reported results is violent in both directions, and Strategy's own figures make the point better than any commentary:

• First half of 2026: an unrealised loss on digital assets of $22,770.8m, a loss from operations of $22,801.4m, a net loss of $20,762.3m, and a loss attributable to common shareholders of $21,392.5m after $630.2m of preferred dividends — $(62.32) a share. The second quarter alone was $(24.45) a share.
• Retained earnings of $6.32bn at 31 December 2025 became an accumulated deficit of $15.20bn six months later. Total stockholders' equity fell from $44,123.5m to $30,886.1m.
• Deferred tax liabilities collapsed from $1,926.5m to $1.4m, producing a $1,921.3m tax benefit inside that loss.
• The comparative quarter, Q2 2025, showed net income of $10,020.8m on a $14,047.5m unrealised gain. Same company, same strategy, opposite sign.

The tax interaction is the sharper risk. Because fair-value accounting raises adjusted financial statement income, it feeds the 15% US corporate alternative minimum tax, which is assessed on that figure. Strategy had flagged a potential multi-billion-dollar CAMT liability from 2026 on more than $27bn of unrealised profits, and warned that it could be forced to sell assets to pay tax on paper gains. Treasury and the IRS issued interim guidance on 30 September 2025 allowing corporations to disregard unrealised gains and losses on digital assets for this purpose. That guidance is interim and not finalised. The risk is suspended, not eliminated.

Separately, on 30 July 2026 the company said it expects distributions on its preferred stock to be treated as a non-taxable return of capital for the foreseeable future, because it expects to have no earnings and profits — while acknowledging that its expectations on earnings and profits may change. A US corporate tax characterisation is not the same thing as its UK tax treatment in your hands, and neither is fixed.

The bear case, and the strongest answer to it

An explainer that only presents one side has not explained anything. Here is the sceptical case as its most credible proponents actually make it, followed by the best good-faith reply.

The case against

• Jim Chanos, July 2025. His argument was about circularity: investors were being asked to value the business not merely at the net market value of the bitcoin, but at a multiple of the change in that net asset value, on the grounds that the balance sheet can be levered. He described the premium as financial gibberish (CoinDesk, 16 July 2025). He was long bitcoin and short the equity throughout, so the trade was a bet on premium compression, not a bet against bitcoin. In fairness to the company: he closed the position after about eleven months, saying that while he still saw room for further mNAV compression, the thesis had largely played out. That is not an open-ended bearish call and should not be presented as one.
• Kerrisdale Capital, March 2024. Kerrisdale, short the equity and long bitcoin, argued that the share price implied a bitcoin price of about $177,000, roughly 2.5 times spot, and that the arrival of low-fee spot ETFs had destroyed the access premium which once justified paying up for a listed bitcoin proxy. It projected roughly a 50% return if the premium reverted to historical averages. The shares fell as much as 14% on publication (CoinDesk, 28 March 2024). We were unable to retrieve the original report, so we are not attributing any more detailed argument to it than the above.
• NYDIG, August 2025. The most structurally careful version. "How DATs Die" identifies premium collapse as the single critical failure mode: the premium derives substantially from narrative around a founder figurehead rather than fundamentals, and can disappear as easily as it appeared. Once it contracts, the company loses its primary growth lever, faces pressure to sell crypto to buy back stock, and pressures the wider market in doing so. NYDIG also flagged saturation — the model works only while investors can make money from it, and exhausting the available investor demand would be an extinction event. It cited ALT5 Sigma pricing a $1.5bn raise at $7.50 a share and closing at $5.94 within days, below net asset value immediately. NYDIG is a bitcoin-focused firm, so this is a critique from inside the industry rather than from a bitcoin sceptic, which arguably strengthens it.

The answer to it

The company's own headline defence is that as at July 2026, MSTR had delivered a 42% annualised return since adopting the bitcoin standard on 10 August 2020, ahead of bitcoin at 32%, the largest US technology stocks at 23% and the S&P 500 at 14%. That claim is arithmetically defensible and should not be waved away. The same leverage that produced the 2026 drawdown produced roughly +346% in 2023 and +318% in 2024. Nor is the structural case trivial: unlike a margined position, common equity has no maturity date and no margin call, the convertible coupons are small, management controls the timing of any sale, and the company survived the 2022 bear market with a leveraged position and a collateralised loan outstanding. The USD Reserve of $5.10bn and USD Cash of $1.59bn at 23 August 2026, and the $1.0bn common repurchase programme and $1.0bn digital credit securities repurchase programme authorised after the June discount, are all deliberate steps to avoid ever being a forced seller. And the position itself moved from about 22% below cost at 30 June 2026 to roughly 4.8% above cost by 23 August 2026, which is a reminder that stress episodes are not the same as outcomes.

The three honest counterpoints to that defence:

1. It is highly start-date dependent. The measurement is anchored on August 2020, before the position was built at any scale. Most current shareholders bought far later.
2. The average cost tells a different story about when the buying happened. The average purchase price across the whole treasury was $75,385 as at 23 August 2026 — reflecting purchases made overwhelmingly at prices far above 2020 levels.
3. A return on the shares is not the same question as value created per share. In the six months to 30 June 2026, claims-adjusted bitcoin per share fell about 16.6% on our calculation, while the reported gross measure rose about 5.6%. A shareholder can have made money on a share whose underlying claim was shrinking, and vice versa. They are two different questions and should not be answered with one number.

What should you watch if you already hold one of these?

Framed as observation, not advice. Nothing in this list tells you to do anything; it tells you what the filings show and where the mechanism becomes visible before it becomes obvious. If you want a decision made about your own money, that is a conversation for a regulated adviser.

• Whether the company is a net buyer or a net seller of the asset. The weekly 8-K states it explicitly. Four consecutive weeks of issuance with no purchases is a different company from the one in the pitch deck.
• The stated use of proceeds. Proceeds going to the asset is the flywheel. Proceeds going to preferred dividends, preferred buybacks and cash reserves is the capital structure being serviced by new shareholders.
• Coins per share, adjusted for senior claims — not the gross headline figure. The two can move in opposite directions for months, as they did in the first half of 2026.
• mNAV, with its definition attached. Basic and enterprise mNAV are different numbers. A quoted multiple without the definition is not usable.
• The preferred liquidation preference and dividend run-rate, against operating gross profit. The gap between $400.7m of preferred dividends in a quarter and $81.6m of gross profit is the structural fact everything else follows from.
• Where the shares sit against convertible conversion prices, and the put calendar. Above the conversion price, the notes are equity. Below it, they are dated cash debt.
• Whether the cash pools are growing or being drawn down. A rising USD Reserve funded by share sales is not the same as a rising USD Reserve funded by operations.
• How the preferred stock is trading against par. A "stable" instrument at a persistent discount, with an escalating coupon and continuous buybacks, is transmitting stress.
• Custody disclosure and any change to it, and index consultation outcomes.
• Date-stamp everything. Strategy's position went from 14.0% below cost at 30 June 2026 to roughly 4.8% above it by 23 August 2026 — under eight weeks. Any claim that a treasury company is "underwater" or "in profit" without a date attached is worthless. So is any coin count taken from an aggregator rather than a filing: during the research for this page, holdings for the same company were quoted variously at 840,447, 842,137 and 847,363, and one aggregator page showed a UK company at 7,500 BTC when its own announcement said 2,712.
• Names and tickers change. MicroStrategy became Strategy Inc; KindlyMD merged with Nakamoto Holdings to form NAKA; Strive acquired Semler Scientific; Twenty One Capital listed via a SPAC. Strategy now has six separately listed securities — the MSTR common plus five perpetual preferred series: STRF, STRC, STRK and STRD on Nasdaq, and STRE in Luxembourg — which coverage routinely conflates. Even sophisticated investors get this wrong: SoftBank paid $999.3m for its Twenty One Capital stake and sold it to Tether in May 2026 at a valuation reported at roughly $711m, an implied loss of about 29%. The transaction terms were not disclosed, so that figure is a press estimate rather than a filed number — but institutional backing is not a substitute for reading the capital structure.

Sources, methodology, and what we could not verify

Company figures on this page are taken from primary SEC filings by Strategy Inc (CIK 0001050446): the Form 10-Q for the quarter ended 30 June 2026, the 8-K exhibit 99.1 of 30 July 2026, the weekly 8-K filings of 3, 10, 17 and 24 August 2026, and the 2022 10-Qs for the Silvergate loan disclosures. The full filing history is on EDGAR. Market prices for MSTR are from stockanalysis.com as at 25 August 2026, corroborated where possible against contemporaneous reporting. Figures are stated in US dollars where the source is a US filing and in sterling where the source is sterling; nothing has been silently converted.

Three categories are kept separate throughout: filed facts, our own arithmetic from filed facts, and third-party estimates. Every calculation on this page that is ours — the residual to common, the leverage factor and sensitivity table, the claims-adjusted coins per share, the four-week per-share dilution, and the approximate convertible coupon burden — is labelled as such at the point it appears.

What we could not verify, stated plainly:

• MSCI's own decision document for the January 2026 consultation outcome. The account here rests on secondary reporting.
• The cumulative or non-cumulative status of each individual preferred series, which we have not restated from the certificates of designations.
• The cost-basis identification method behind the 2026 bitcoin disposals. The implied original cost of the coins sold sits far above the portfolio average, which is consistent with selling highest-cost lots first, but the filing does not state the method.
• Strategy's split of bitcoin between its three custodians, which is not disclosed, so actual single-custodian concentration cannot be assessed.
• The Smarter Web Company's share price history, where sources are irreconcilable in a way that may reflect a share consolidation. No price history for it is quoted here.
• Metaplanet's exact bitcoin holdings, which are contested between sources.
• Whether shares in a UK-listed bitcoin treasury company are in fact eligible for a stocks and shares ISA. This is flagged as our inference in the relevant section.
• Realised and implied volatility figures for MSTR against bitcoin, which were inconsistent and largely undated across sources. This is why the page uses drawdowns and annual returns, which are far better evidenced, rather than any volatility percentage.

Our methodology page and corrections policy set out how we source and correct. This is an educational explainer about how an instrument works and how it fails. It is not personal advice, not a recommendation, and not a view on any price. If you are weighing a decision that matters, take regulated advice.

Related digital asset and risk guides

• Crypto CGT in the UK for 2026/27
• Crypto tax calculator
• UK ETF guide
• Scam and fraud defence centre
• How to start investing in the UK
• Complete UK ISA guide
• Complete UK CGT guide