Invest1 publisher3 min readPublished
Marathon's $98.6M September Buy Replaces About 6% of the Bitcoin It Sold in Q1
Lookonchain spotted 1,292 coins moving to Marathon through FalconX on September 15. With a 1,000-coin purchase in June, the miner has bought back 2,292 of the 20,880 it sold for $1.5 billion, at a higher price.
The Investor · Invest desk
What happened
- Marathon Digital Holdings bought 1,292 BTC for roughly $98.6 million through the institutional prime broker FalconX, and Lookonchain identified the September 15 transaction before the company announced it.
- The miner had spent the first quarter of 2026 selling about 20,880 BTC for around $1.5 billion, putting the money into operating expenses, liquidity management and an AI and high-performance computing push.
- Holdings went from roughly 53,822 BTC in late 2025 to approximately 35,303 BTC by March 2026.
- This was the second purchase of the year, following 1,000 BTC bought through the same FalconX channel in June 2026.
- Early 2026 filings valued the treasury between $2.4 billion and $2.7 billion, a level at which MARA occasionally trades the second-largest public holder title with MicroStrategy.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint Replacing the remaining 18,588 coins would cost about $1.42 billion at September's price. A full rebuild is a capital-raising exercise the size of the original sale programme.
- decision Every coin bought after this one has to come from either mining output or the cash pool that funds the compute buildout, and at 2,292 coins against 2,361 of implied production the company has not yet had to choose.
- exposure Off-exchange execution hides the buying from the order book but not from the ledger, so an analytics account sets the disclosure timetable for Marathon's treasury decisions.
- contradiction The sales tally and the balance decline differ by about 2,361 coins, so anyone treating the sale of a third of the treasury and a third-sized fall in holdings as the same statement is off by a quarter's production.
The first-quarter sales averaged roughly $71,839 a coin, $1.5 billion across 20,880 coins [5][1]. September's purchase went through at about $76,347 [1]. That is $4,508 a coin more, 6.3% [2]. Repurchasing the whole sold position at that price would cost about $1.594 billion against the $1.5 billion raised, so the completed round trip would put about $94 million more out than came in [4].
Marathon is nowhere near completing it. The 1,292 coins bought on September 15 and the 1,000 bought through the same FalconX channel in June come to 2,292, or 11% of what went out the door [4][8][3]. The $98.6 million spent in September is 6.6% of the first-quarter proceeds [8], and 1,292 coins is 3.7% of the 35,303 the company held in March [7].
The sales figure and the holdings figures track two different quantities. A fall from 53,822 coins to 35,303 is 18,519 [7][5], against 20,880 sold, which leaves about 2,361 coins arriving from mining or elsewhere over the same months [6]. The two purchases together, 2,292 coins, are 97% of that implied quarterly production [11].
Cryptobriefing reported that the $1.5 billion went to operating expenses, liquidity management and a strategic push into artificial intelligence and high-performance computing infrastructure [6]. The report does not split the total between those three uses. It describes FalconX as a way for a large buyer to source liquidity across several counterparties at once and keep slippage low [11], and it treats the June purchase as a sign the sell-heavy posture was transitional [12].
Several readings fit the record. It could be a rebuild toward a revised target funded as cash allows; it could be opportunistic buying near a price the company happens to like, with no target at all; or it could be maintenance at roughly the rate the machines produce, which is what 2,292 against 2,361 looks like to me. I lean to the third, and it matters for how you read the compute spending, because buying inside your own output does not compete with a capital budget.
On the broader claim, that miners as a class are rebuilding reserves after funding AI buildouts, the evidence here covers one company. Cryptobriefing lists whether other major miners take this as a cue among its open questions, alongside a revised treasury target and the maturing of the AI and HPC bets [13]. Early 2026 filings put Marathon's treasury between $2.4 billion and $2.7 billion, so $98.6 million is 3.7% to 4.1% of it [9][9]. I would change my mind on a disclosed target with a number attached, or on a quarter in which purchases run well past the roughly 2,361 coins of implied output.
What to watch
- A disclosed revised treasury target with a coin count or dollar figure attached to it.
- A quarter in which MARA's purchases run well above the roughly 2,361 coins of implied production.
- A second large listed miner buying back coins it sold. That would turn this from one company's cash management into a sector pattern.