MARA Holdings published its second quarter 2026 report on 6 August 2026, and we are writing about it today.
Every figure here was verified against the Form 8-K filed with the SEC under Item 2.02, accession number 0001507605-26-000020, and against the Form 10-Q filed the same day, accession number 0001507605-26-000022. Note that the exhibit order here is inverted from the usual convention: the substantive document is Exhibit 99.1, the shareholder letter, while Exhibit 99.2 is a short release that only announces the webcast. Even so, errors, inaccuracies or omissions are possible, and the figures may change after publication. Spotted something that looks wrong? Write to me and I will correct it.
What the Company Does
MARA mines bitcoin. In plain language: the bitcoin network issues new coins as a reward to whoever first solves a computational puzzle. Purpose-built machines around the world guess answers at enormous speed, and whoever lands the answer first "wins a block" and receives the coins. That is the whole activity: computers, electricity, and a statistical lottery whose odds are simply your share of all the guessing on the network.
The industry's central metric is hashrate. It is measured in EH/s, exahashes per second, meaning a quintillion guesses every second. The larger your rate, the larger your share of the new coins. This quarter MARA won 5.9% of all available miner rewards, against 5.5% in the prior quarter.
Revenue follows directly from that: the value of the bitcoin mined. There are no customers here, no contracts and no backlog. Revenue is a function of two things only: how many coins you produced, and at what price.
What the company is trying to build alongside it is a second business: leasing the infrastructure itself, the power and the data centres, to heavy compute customers and above all to AI customers. The company calls this Digital Infrastructure, and it acquired Exaion, a European compute company, to that end. In this report, as we will see, that is still a story rather than a revenue line.
The Quarter
| Metric | Q2 2026 | Q2 2025 |
|---|---|---|
| Revenue | $174.9 million | $238.5 million |
| Net income (loss) | Loss of $611.3 million | Income of $808.2 million |
| Change in fair value of bitcoin | Loss of $343.0 million | Gain of $1,192.6 million |
| Basic earnings (loss) per share | Negative $1.60 | $2.29 |
| Adjusted EBITDA | Negative $360.9 million | Positive $1,245.4 million |
| Energized hashrate | 70.3 EH/s | 57.4 EH/s |
| Bitcoin produced | 2,422 | 2,358 |
| Blocks won | 700 | 694 |
| Cost per petahash per day | $27.7 | $28.7 |
| Purchased energy cost per bitcoin | $38,690 | $33,735 |
| General and administrative | $114.7 million | $92.9 million |
| Depreciation and amortization | $174.7 million | $161.7 million |
On the revenue decline the company gives an orderly bridge: roughly $65.9 million was erased by a 28% lower average bitcoin price, against which came a $7.2 million addition from higher production, and a further $4.9 million decline in other revenues, partly because hosting services for customers were eliminated. There is no separate AI or HPC revenue line anywhere in the report.
The Angle: The Loss Is the Bitcoin Price. And So Was Last Year's Profit
The line that explains everything is called change in fair value of digital assets. Since a new accounting standard took effect, a company holding bitcoin remarks it to market value every quarter, and the difference flows straight into the income statement. Even when not a single coin has been sold.
This quarter that line produced a $343.0 million loss. It is split across two places in the filing: $249.559 million inside operating costs, plus $93.456 million in other income, on digital assets receivable. A year ago the same line produced a gain of $1,192.6 million, made up of an $846.027 million operating gain plus $346.547 million in other income.
The year-over-year swing in that single line is $1.536 billion.
Stop on that number for a moment. The entire net loss is $611.3 million. This non-cash line is larger than the whole loss, and the swing in it is larger than the entire collapse in earnings. In other words: you do not need any operational deterioration to explain the move from profit to loss. It is already explained, with room to spare, by the bitcoin price, which fell 45% per the figure stated in the filing. The spot price used to mark the holding at 30 June 2026 was $58,524.
And it works exactly the same way in the other direction. The $808.2 million "profit" a year ago was not operating profit. It was mostly a $1.19 billion revaluation gain on bitcoin the company happened to hold while the price rose. Anyone who read that report as proof the business prints money was making precisely the mistake being made today by anyone reading this one as an operational blow-up.
This is the same accounting mechanism I wrote about at MicroStrategy. The bottom line of a company holding bitcoin on its balance sheet is, in practice, a leveraged bet on the bitcoin price rather than a measurement of the business.
An asterisk that actually works in the company's favour, which is why it is easy to miss.
MARA's adjusted EBITDA is negative $360.9 million. But the definition the company itself chose does not strip out the bitcoin revaluation. It adds back stock-based compensation, derivative fair value, impairment, restructuring, acquisition and integration costs, a litigation settlement, debt extinguishment gains and investment gains. The single largest item, the $343.0 million bitcoin loss, it leaves inside.
That is unusual relative to peers, and it works against the company's own headline. Back that line out and the measure falls to roughly negative $18 million. Which is to say the "clean" business was close to breakeven, far better than the non-GAAP number signals. This is the rare case where the adjusted metric makes a company look worse than it is.
The AI Infrastructure Pivot: Zero Disclosed Revenue
This is the part to be most careful with, because it is the story being sold to investors.
MARA's revenue this quarter is one hundred percent mining. There is no AI segment, no disclosed signed compute contract, no announced customer, and not one revenue line attributable to infrastructure.
And the two central assets the thesis rests on are both unclosed and gated by regulators: the Long Ridge acquisition awaits approval from FERC, the federal power regulator; and the 2 gigawatt site in Matagorda County, Texas is subject to approvals from ERCOT, the Texas grid operator, and to interconnection approval. The figure the company presents, a power portfolio of up to 4.8 gigawatts, is a pro-forma number that assumes both deals close. It is not existing capacity.
The company does not hide this. The shareholder letter concedes that the second half "is about execution", and describes lease discussions as only being "advanced".
Chairman and Chief Executive Fred Thiel put the thesis in one line in the letter:
"Capital is abundant. Power-ready sites are scarce."
Meaning: capital is plentiful, sites already connected to power are the scarce good. That is an entirely reasonable thesis. It simply has not yet become revenue.
And in the same letter, Thiel wrote this as well:
"Ultimately, our shareholders should judge us not by our vision, but by our execution."
It is hard to put it better than he did, so I will simply take him at his word.
The bitcoin treasury is being drained and encumbered at the same time, and those are two different things happening together.
Drained: the holding fell 29% year over year to 35,577 bitcoin. This quarter the company sold 2,213 bitcoin at an average of $73,078, against 2,422 produced. That is roughly 91% of everything it mined. Not one bitcoin was purchased.
Encumbered: of what remains, 9,270 bitcoin, 26% of the holding, are already loaned out or pledged as collateral. 4,742 loaned and 4,528 pledged. That leaves 26,307 bitcoin genuinely unrestricted. The lending did generate roughly $4.3 million of interest income in the quarter.
And after quarter-end the company added $600 million of new bitcoin-backed credit facilities with Coinbase and Two Prime, at a 7.56% weighted average cost, refinancing an existing $150 million facility into them. Management presents this as financing that does not dilute shareholders, and that is true. But it also converts the bitcoin treasury into collateral against a falling asset. The very same asset whose fall created the loss in this report.
Here is how Chief Financial Officer Salman Khan framed it in the capital allocation section:
"We have consistently said that bitcoin is one of MARA's most strategic assets."
That sentence is true in both directions. A strategic asset is also an asset serving as collateral, and an asset from which 91% of production is sold to fund operations.
What Sits Underneath the Headline
The Diluted EPS Comparison Is Not Fair
The company shows a diluted loss of $1.60 per share against diluted earnings of $1.84 a year ago. Do not use that comparison. In a loss quarter, potential shares drop out of the calculation because they would shrink the loss per share, which accounting does not allow. That removed 82,691,847 potential shares, including 49.4 million convertible note shares, 12.4 million restricted stock units and 21.0 million performance units.
The result: the diluted share count this quarter equals the basic count, 381,565,856, against 440,912,159 diluted shares a year ago. The honest comparison is basic to basic: negative $1.60 against $2.29. The weighted average basic share count rose 8.1% year over year.
Operating Discipline Is Mixed, Not Simply Improving
The number management leads with is cost per petahash per day: $27.7, a 4% improvement, and a 27% improvement across the past nine quarters. That is a real metric and a real improvement. It is also the narrowest one in the report.
Beside it: purchased energy cost per bitcoin rose 15% to $38,690, on higher power prices and because global network difficulty grew faster than hashrate. Total purchased energy costs rose 17% to $48.750 million. Third-party hosting and other energy costs were essentially flat at $69.156 million.
And general and administrative expenses: $114.7 million on a GAAP basis, up 23%. Excluding stock-based compensation they jumped 73%, from $40.1 million to $69.5 million. Even stripping out $15.4 million of acquisition and integration costs and a $10.2 million settlement of a patent dispute, the underlying base still rose to roughly $43.9 million.
Depreciation Hides a Warning Sign
Depreciation and amortization rose 15% to $174.7 million, and inside it sits $28.1 million of accelerated depreciation on certain mining rigs. Accelerated depreciation means the company shortened the expected life of equipment it already owns. That is an early indicator of fleet obsolescence, and it is disclosed without any detail on how much of the fleet is affected.
The Sequential Direction of Hashrate Is the Opposite of the Headline
The headline is 70.3 EH/s, up 22% year over year. But the prior quarter stood at 72.2 EH/s. That is a 3% decline in the quarter. Management notes adverse weather events that raised power costs versus the prior quarter. In this industry the sequential direction is the live one.
The Balance Sheet Contracted Sharply, and It Is Barely Discussed
Within six months: total assets fell from $7.287 billion to $4.349 billion. Equity attributable to MARA fell from $3.472 billion to $1.661 billion, less than half. The accumulated deficit widened from $1.338 billion to $3.207 billion. The current portion of notes payable jumped to $335.4 million from $47.8 million, concentrating near-term maturities. Cash fell to $421.3 million from $547.1 million. Cash plus bitcoin, including the loaned and pledged coins, stands at roughly $2.5 billion.
On the issuance side, by contrast, it was quiet: shares issued and outstanding rose only 0.6% in the half, from 379,464,892 to 381,888,004.
Three Smaller Notes Worth Knowing
First, the derivative in this report is not a warrant. Per the 10-Q, MARA's "change in fair value of derivative instrument" is valued by discounted cash flow off electricity forward curves. It is a power hedge, not an equity-linked instrument. It came to a $1.769 million expense in the quarter, though $42.814 million across the half. Anyone who followed the warrant revaluation story at other miners cannot import it here.
Second, the half-year figures contain a one-off that does not repeat in the quarter: a $70.557 million net gain on extinguishment of debt. The six-month loss is $1,873.7 million against $274.8 million of income a year ago, and six-month revenue is $349.5 million against $452.4 million. The half also carries $47.6 million of restructuring costs, almost all of it in the first quarter. Only $1.753 million landed in this one.
Third, the Exaion consolidation brought $89.7 million of goodwill and $40.5 million of intangibles onto the balance sheet, from essentially nothing at year-end, alongside an $82.5 million redeemable noncontrolling interest and a $25.3 million contingent consideration liability. Goodwill in a sector this volatile is a natural candidate for future impairment. The company already took a $26.3 million impairment in the year-ago quarter.
And a note on timing: two directors, Barbara Humpton and Georges Antoun, resigned effective 31 July 2026, disclosed in a separate filing on 4 August, two days before earnings. Both were stated as personal reasons, with no disagreement over operations or policies, and both seats were filled on 1 August. Almost certainly entirely routine, and properly disclosed. I note only the timing.
Guidance
And here it has to be said plainly: there is no numeric guidance. Not for revenue, not for EBITDA, not for hashrate and not for capital expenditure, not for the third quarter, the fourth, or the full year. The absence of guidance is itself a data point.
What was said is entirely qualitative:
- The company expects the quarterly general and administrative run-rate, excluding stock-based compensation and acquisition and integration costs, to keep trending lower as the savings are realised
- Over the coming months it expects to complete the Long Ridge acquisition, advance lease discussions across the Digital Infrastructure portfolio, expand Exaion internationally and commercialise technology initiatives
- It intends to keep monetising bitcoin opportunistically to fund operations and capital projects
- Long Ridge and Matagorda together would take the power portfolio to up to 4.8 gigawatts, subject to FERC, ERCOT and interconnection approvals
- Third-party hosting arrangements begin expiring in the third quarter of 2027 and conclude in the first quarter of 2028, which is expected to eliminate hosting costs and improve cost per kilowatt hour
- An Investor Day is planned for later this year
Not one of those items carries a date or a dollar figure.
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
The big number in this report is not the loss. It is the swing. A $1.54 billion move in a single line through which not one dollar of cash passed, explaining more than one hundred percent of the move from profit to loss. Anyone describing this report as an operational collapse at MARA simply did not open the filing.
Which is exactly why I insist it works in both directions. The $808 million profit a year ago was manufactured in precisely the same way. If you bought the story then, you have to accept the number now. You do not get to choose which quarter the accounting is real in.
What I do look at is the business underneath, and there the picture is mixed rather than terrible. Strip out the bitcoin revaluation and adjusted EBITDA is close to breakeven, which is better than the headline. But energy cost per bitcoin rose 15%, general and administrative expenses excluding stock compensation jumped 73%, and hashrate fell sequentially. The one metric improving nicely is also the narrowest one.
The gap that troubles me most is between the story and the numbers. The company is sold today as an AI infrastructure player, yet this report has no AI revenue line, no signed lease, and two central assets waiting on regulators. A 4.8 gigawatt portfolio is an aspiration, not capacity. Thiel himself wrote that shareholders should judge the company by its execution rather than its vision, and I intend to do exactly that.
So four things I am watching next quarter: whether FERC has approved Long Ridge; whether a disclosed revenue line from leasing infrastructure finally appears; where hashrate goes sequentially rather than year over year; and how much of the remaining bitcoin is pledged. That last one worries me more than the rest, because it turns a volatile asset into collateral, and volatile collateral behaves badly at exactly the moment it is least convenient.
