MARA Holdings disposed of nearly a third of its bitcoin treasury in Q1 to retire zero-coupon convertibles and fund a pivot toward AI infrastructure — and its 10-Q no longer describes the firm primarily as a miner.
MARA Holdings sold 20,880 bitcoin in the first quarter for roughly $1.5 billion — nearly a third of its treasury — and used $1 billion of the proceeds to repurchase zero-coupon convertible notes due in 2030 and 2031. The Florida-based miner reported the disposal in its Q1 10-Q on May 12 alongside a $1.26 billion net loss, an 18 per cent year-on-year revenue drop to $175 million, and a self-description that no longer leads with mining.
The 10-Q calls MARA "a digital infrastructure company built to convert energy into high-value compute workloads." Bitcoin mining now sits alongside artificial intelligence and high-performance computing as one of three equal priorities. It is the most explicit reframing any major listed miner has put into an SEC filing this cycle, and it lands a week after the company committed $1.5 billion to acquire the Long Ridge Energy & Power campus in Ohio — a 505 MW gas plant whose load was always going to outgrow what hashboards alone could absorb.
The sale itself was opportunistic. MARA disposed of roughly 15,133 BTC near the end of March specifically to retire $1 billion of convertibles — about $367.5 million of the 2030 paper and $633.4 million of the 2031 paper — at a discount to par. That trade compressed reported leverage from $3.3 billion to $2.3 billion in a quarter when bitcoin slid back toward $75,000. CEO Fred Thiel had spent two earnings cycles defending the strategy of holding every coin mined; the Q1 release abandons that framing without quite admitting it.
What is left looks like a different company. MARA still holds 35,303 BTC, worth about $2.84 billion at this morning's spot, which makes it the fourth-largest public bitcoin treasury rather than the second. It will continue to mine — Thiel was explicit on the call that hashprice still clears variable costs at scale — but the company has no plans to buy new ASIC machines. The implication is that the hashrate footprint MARA ends 2026 with is roughly the one it has today, and that the marginal megawatt at every site it acquires from here will be priced against AI rents, not block rewards.
The market read this two ways. The shares fell after hours on the headline loss, which was nearly 2.4 times the Q1 2025 figure, and a substantial part of that came from unrealised bitcoin markdowns. But the spread between MARA's enterprise value and its net asset value has narrowed materially since the 10-Q dropped, suggesting at least some institutional desks are willing to underwrite the AI thesis at a higher multiple than they were giving the mining business in isolation.
The wider miner cohort is moving in the same direction with less precision. CoreWeave's earlier pivot is the obvious template, but among the public miners that actually mine for themselves — Riot, CleanSpark, Hut 8, Iris Energy — every Q1 disclosure has contained more language about compute and less about hashrate guidance than the previous one. Riot still calls itself a bitcoin company. Hut 8 has its AI venture spun out as American Bitcoin. CleanSpark hedges. MARA's filing is the cleanest break with the original thesis.
The risk is straightforward. A 90 per cent reallocation of non-hosted mining capacity toward AI workloads — the upper-bound figure executives floated on the call — assumes long-tenor hyperscaler contracts at margins miners have never operated under, against competitors with twenty years of data centre design experience and balance sheets denominated in trillions. Power, which was always MARA's edge, is no longer scarce in the way it was in 2021. The same Texas substations the miners colonised are now bid by Microsoft, Amazon, and a queue of GPU-rental startups that did not exist eighteen months ago. MARA's pitch is that its sites came online when nobody else wanted them, and that head-start has years of value. The hyperscalers' counter is that they will simply pay more.
Convertibles maturing in 2030 and 2031 were the cheapest debt in MARA's stack — zero-coupon, deep in the money on conversion terms set in a cycle when the stock traded in the $30s. Buying them back at a discount to par sounds clever, but it removes the cleanest financing tool the company had for funding any large AI capex without diluting equity. The Ohio acquisition will need other forms of debt; the next $1 billion will be more expensive than the $1 billion that just came off the balance sheet.
The pivot also sets the company against a different kind of competitor. Coinbase cut 14 per cent of its workforce a fortnight ago and pinned the decision on AI. Strategy continues to add to its 818,000-coin stack at every price level. The MARA position now is neither a pure-play miner nor a pure-play treasury, and the equity will be priced by whichever cohort of institutional buyer is willing to underwrite both halves at once.
For now, MARA has done the rare thing of executing a pivot in public without the share-price reaction destroying its option value. The company that bought every coin it could mine in 2024 has just sold a third of its hoard and rewritten its self-description in the same filing. The remaining 35,303 bitcoin are still on the books. The next quarterly report will say whether they stay there.
