Riot Platforms (RIOT) and Strategy, Inc. (MSTR) both sold Bitcoin (BTC) in the first week of August and both disclosed it on 10 Aug. Riot spent the proceeds on data centres that will pay rent from December 2027, and its shares rose 25% after hours. Strategy spent its proceeds buying back its own preferred stock. Two companies are pointed at opposite ends of the same trade, and the difference is what a Bitcoin balance sheet is actually for.
Riot, a crypto miner and digital infrastructure company based in Castle Rock, Colorado, sells its entire monthly Bitcoin production and has drawn down its Bitcoin inventory. Strategy, the Tysons Corner, Virginia software company that holds Bitcoin as its principal treasury asset, has sold 1,690 BTC between 3 and 9 Aug for $109mn.
The number that explains why Riot is doing any of this is not in its presentation. Riot's mining operation drew 1.79bn kilowatt-hours in the quarter, an average of 818 megawatts across the period. Annualized, that power produced about $455mn of mining revenue and, on Riot's own preferred measure, about $138mn of mining gross profit. That works out at roughly $560k of revenue and $170k of profit for each megawatt drawn, each year. Its two signed data-centre leases produce $520mn of average annual revenue and $416mn to $462mn of net operating income across 241 megawatts of critical IT capacity. Converting critical IT to total facility power at the ratio Riot itself publishes for Corsicana, where 1 gigawatt of utility power supports 756 megawatts of critical IT, that is about 319 megawatts of facility draw, or roughly $1.6mn of revenue and $1.3mn to $1.4mn of net operating income per megawatt.
Same power, same sites, same interconnection. Leased to AI tenants, it earns about three times the revenue and roughly eight times the profit that mining the largest cryptocurrency earns. That is the entire strategy in one ratio, and it is why the quarterly loss did not matter to anybody on the call.
The comparison is deliberately generous to mining. Dividing revenue by average draw rather than installed capacity flatters the per-megawatt figure because Riot curtails and earns credits for doing so. Measured against capacity, mining looks worse.
The market bought the lease, not the quarter
Riot shares had already fallen 5.5% through the session on 10 Aug, closing at $19.37 against $20.50 on 7 Aug. The results and the lease landed at 20:00UTC, and the stock opened the following hour at $19.38 and touched $22.00 within it, ending the hour at $21.45.
It kept climbing through the earnings call, which began at 20:30UTC, reaching $24.29 in the hour to 00:00UTC on 11 Aug, up 25% on the regular-session close. By 10:00UTC on 11 Aug, it had eased to $23.00 in pre-market trading, still 19% above the 10 Aug close and 12% above where it started the week. Prices are hourly, derived from TradingView.
So the repricing was not a reflex to a headline number. It built over four hours, across the call, on the detail.
Mining now costs more than it produces
The per-megawatt gap explains the direction of travel. Riot's income statement explains the urgency.
The company's reconciliation tables show the cost to mine one bitcoin including miner depreciation was $90,631 in the quarter, against a production value per bitcoin of $71,667. That is 126.5% of what one bitcoin was worth. The first quarter was 126.7%. The fourth quarter of 2025 was 106.6%. On that fully costed basis, the mining segment lost $40mn in the quarter and $91mn across the half.
Riot disputes the relevance of that measure in a footnote to the same table, arguing it excludes depreciation when judging whether running miners is economic because depreciation is non-cash and cannot be avoided by curtailing. On a cash basis, mining earned a 30% gross margin. Both positions are defensible, and the per-megawatt comparison above does not depend on resolving them: even on Riot's own cash measure, the power earns eight times more under a lease.
What has changed is not costs. Cash cost to mine rose 1.9% year-to-year, from $48,992 to $49,912, and an all-in cost of power of 3.6 cents per kilowatt-hour is genuinely low. Production value per bitcoin fell 27%, from $98,800 to $71,667. Riot mined 1,587 coins against 1,426 a year earlier and made $27mn less doing it, with mining revenue down 19% to $114mn.
CEO Jason Les described where mining now sits without dressing it up. Riot will "continue to utilize Bitcoin mining to monetize that available capacity until we have the right lease or set of leases." Mining is a yield strategy for megawatts nobody has leased yet.
The accounting says the same. Riot took a $28mn impairment on mining-related construction at Rockdale, which it attributed directly to repurposing that capacity for data centres. It paid to demolish the old use to make room for the new one. Keel Infrastructure shut its US mining operations entirely last week on the same logic.
What the cash is buying
Riot's Bitcoin holdings, including restricted coin, fell from a carrying value of $1.58bn at the end of 2025 to $666mn at 30 Jun, based on a closing price of $58,527. Cash and restricted cash went the other way, from $310mn to $549mn.
CFO Jason Chung told the call that Riot sells all of its monthly Bitcoin production, has "meaningfully sold down" its inventory, and that those proceeds remain the primary funding source for the equity component of data-centre capital expenditure. Riot issued no common equity during the quarter. Of the 11,380 coins left, 5,821 are pledged against a $200mn credit facility.
The Bitcoin is being consumed as raw material. It becomes switchgear, substations and, eventually, a rent roll.
Strategy's week looked nothing like that
Strategy's 8-K for the same period shows 1,690 BTC sold at an average $64,262, against an average purchase price across its holdings of $75,385. It sold roughly 15% below its own cost basis. The $109mn went, dollar for dollar, into repurchasing 1.15mn STRC preferred shares under the digital credit securities repurchase programme announced on 29 Jun. Some $785mn of that authorization remains.
In the same week, Strategy sold 6.59mn MSTR shares through its at-the-market programme for $653mn of net proceeds, of which $650mn went straight into what it calls the USD Reserve. That reserve stood at $4.65bn as at 9 Aug, against expected annual preferred dividend and interest payments of about $1.76bn disclosed in the 10-Q filed on 3 Aug. The board's floor is 12 months of coverage.
The pattern is now five disclosed sales in 2026, totalling about 6,948 BTC for roughly $432mn. After a 32-coin sale in late May at about $77,135, the company sold 1,363 BTC at $59,256 in late June, 2,225 at $60,773 in early July, 1,638 at $63,957 in the week to 2 Aug and 1,690 at $64,262 in the week to 9 Aug. The realized price has risen through the sequence and has not once come close to the cost basis. The Bitcoin monetization programme announced on 29 Jun permits up to $1.25bn of sales, so roughly a third has been used, and Strategy has disclosed no purchase since June.
One figure needs correcting. Several outlets have carried a USD Reserve of $4.75bn this week. The 8-K says $4.65bn.
Strategy still holds 840,447 BTC and bought heavily in the first half. CEO Phong Le told CNBC on 3 Aug that the company is not a forced seller, and on the arithmetic of a $4.65bn reserve against a $1.76bn annual bill he has a case. But every dollar raised that week, from Bitcoin and from shareholders alike, went to servicing or shrinking Strategy's own capital structure. Sandmark's report on the second consecutive week of sales funding the $1.1bn STRC burden set out the mechanism.
The distinction that matters
Both companies have concluded that Bitcoin is more useful converted than held. That is the shared and underreported finding of this reporting week, and it cuts against the treasury doctrine of the past two years.
Where they part is what the conversion funds. Riot is turning a megawatt earning $170k into a megawatt earning $1.3mn. Strategy is turning Bitcoin and freshly issued equity into coverage on dividends generated by the instruments it used to buy the Bitcoin. One company is reallocating an asset to a better use. The other is defending a structure.
For a holder of either security, that is the question. Riot's Bitcoin sales stop when the data centres pay. Strategy's preferred obligations do not stop, which is what perpetual means.
Rent does not start until December 2027
Riot's version has a hole in the middle of it. Rent commences with the first 96 MW in December 2027, and the remaining 95 MW follows in June 2028. Capital expenditure on the project runs to $2.1bn to $2.3bn, and Chung said spending ramps through the second half of this year, peaks in the first half of 2027, and combined with the AMD build-out is heaviest around the second and third quarters of 2027.
That is roughly six quarters of maximum outlay before the tenant pays anything, funded from a Bitcoin stack that has more than halved in six months and a mining business whose per-megawatt economics are the reason the outlay exists. Recurring operating lease revenue in the quarter was $4.9mn.
The financing plan is coherent on its own terms. A term loan of about $180mn against delivered AMD capacity, roughly twice the $90mn already spent, yields about $90mn of net new capital. Debt at 80% to 90% of cost on the Rockdale project covers $1.7bn to $2.1bn, leaving $210mn to $460mn of equity, cut to $30mn to $280mn after the AMD proceeds, against $1.2bn of liquidity. Every step depends on financing closing at expected terms.
One of those terms is not yet agreed. The tenant is described throughout as "one of the world's leading frontier AI labs" and is never named; Bloomberg has reported, citing people familiar with the situation, that it is Anthropic. Riot has a $573mn interim facility from Morgan Stanley, which Chung said is fully secured by the tenant, while permanent financing waits on an investment-grade credit backstop that is "in finalization". He declined to name the provider or the terms. The bridge exists because the takeout does not.
What to watch
Corsicana is the larger prize and the looser commitment. The full 1 GW site, capable of 756 MW of critical IT capacity, is under a non-binding letter of intent to a single tenant. Les said a full-site lease would generate in excess of $1bn of annual rent at full deployment, and volunteered that Riot has never previously disclosed a letter of intent, having had one before every deal it has announced. Choosing to disclose this one is itself information.
The engineering arm is the other thing to track. ESS Metron and E4A Solutions produced $37mn of revenue against $11mn a year earlier, at a gross margin of 27.5% against 6.8%, with a backlog of $177mn that Riot says is 90% data centre. ESS Metron makes the low and medium voltage switchgear and power distribution units that are among the most constrained items in the data-centre supply chain, which is why Riot could sign a lease in August and commit to 96 MW sixteen months later.
Riot's own comparison table, sourced from FactSet as at 3 Aug with balance-sheet inputs as at 31 Mar, puts it at $3.7mn of enterprise value per 2027 available megawatt against $8.9mn for Cipher Digital (CIFR) and $8.6mn for Hut 8 (HUT). It is a company-produced comp set with stale inputs, and it tells you what management thinks the equity story is: the megawatts are contracted, the mining is a legacy line, and the multiple has not caught up.
The four-hour repricing on 10 Aug suggests some of the market agrees. Between now and December 2027, though, Riot is a company with a shrinking Bitcoin treasury, an unsigned credit backstop, a peak capital expenditure schedule and 818 megawatts still earning mining economics. It has 18 months to move them.
Sandmark approached Riot for further information regarding facility power, the credit backstop and engineering, but had yet to receive at answer at the time of publication.
