The numbers MARA Holdings released on February 26th were always going to be rough. Bitcoin fell roughly 30% during Q4 2025, and MARA holds a lot of Bitcoin. What nobody quite anticipated was just how bad the accounting would look once fair-value adjustments hit the income statement.
A $1.71 billion net loss for the quarter. A year ago, the same period produced $528.3 million in net income. That reversal, more than $2.2 billion in the wrong direction over twelve months, is the kind of swing that forces a company to ask serious questions about whether its current structure makes sense.
MARA is asking those questions. The answers it’s arriving at involve data centers, AI infrastructure, and a French energy company. More on that below.
What Drove the Loss
The headline figure is large enough to obscure what actually happened, so it’s worth being precise. The $1.71 billion loss was not primarily an operational failure. It was an accounting consequence of Bitcoin’s price movement.
Bitcoin opened Q4 2025 at approximately $114,300 and closed it at roughly $88,800, a 22% decline over the quarter. Under current accounting standards, MARA marks its Bitcoin holdings to fair value, meaning that price drop generated a $1.5 billion negative adjustment that flows directly through the income statement. Remove that adjustment and the picture, while still weak, looks materially different.
That said, the operational numbers weren’t clean either. Revenue came in at $202.3 million for the quarter, down 6% year-over-year and meaningfully below the analyst consensus of around $251 million. The cost of producing each Bitcoin, measuring purchased energy costs, rose 54% year-over-year to $48,611 per coin. Hashrate grew 25% to 66.4 EH/s, but actual production fell to 2,011 BTC from 2,144 in Q3 as network difficulty climbed faster than MARA’s capacity additions.
EPS came in at a loss of $4.52 per share. The consensus had penciled in a $0.45 loss. The miss was not close.
The Treasury Position
Despite the quarterly carnage, MARA ended 2025 holding 53,822 BTC, a position valued at approximately $4.7 billion at year-end prices. That’s a substantial asset base, and the company has been working to generate yield from it rather than simply holding.
About 28% of holdings, 15,315 BTC, were loaned or pledged as collateral during 2025, generating $32.1 million in interest income over the year. It’s a meaningful revenue stream given the size of the portfolio, though it introduces counterparty risk that becomes more relevant when prices move sharply.
One notable operational change in Q4: MARA did not use its at-the-market equity program for the first time since 2022, opting instead to sell Bitcoin to fund operations rather than dilute shareholders. Whether that reflects a philosophical shift or simply a tactical decision given market conditions, it signals something about how management is thinking about capital allocation heading into 2026.
The AI Pivot
The more consequential announcements in the shareholder letter had nothing to do with Bitcoin mining. MARA is moving aggressively into AI and high-performance computing infrastructure, a pivot that, if it plays out, would fundamentally change what kind of company this is.
The first piece is a joint venture with Starwood Digital Ventures to develop data centers. The target is 1 gigawatt of near-term IT capacity, with a stated pathway to 2.5 GW, numbers that put the ambition firmly in the category of large-scale infrastructure buildout rather than a side project.
The second piece is an acquisition: MARA took a 64% stake in Exaion, a subsidiary of EDF, the French state-owned energy giant. Exaion operates sovereign-grade AI and private cloud infrastructure, giving MARA both technical capabilities and an entry point into European institutional markets that would have been inaccessible through mining alone.
The logic behind both moves is straightforward. Bitcoin’s fair-value accounting creates earnings volatility that is essentially impossible to manage operationally, when the asset moves 30% in a quarter, the income statement moves with it regardless of how efficiently the mining business runs. AI infrastructure, by contrast, generates contracted revenue streams that don’t reprice every time macro sentiment shifts.
Where That Leaves MARA
The company exits Q4 2025 with a battered income statement, a large Bitcoin treasury, rising production costs, and two significant bets on infrastructure businesses that didn’t exist in its portfolio twelve months ago.
The Starwood and Exaion announcements are early stage, capacity targets and strategic rationale are clearer than execution timelines right now. What they represent is a company that looked at Q4’s numbers and concluded that doing more of the same wasn’t a viable answer.
Whether the AI pivot generates the stable cash flows MARA is looking for depends on execution quality and market timing that can’t be assessed from a shareholder letter. The directional logic, at least, is hard to argue with.
