MARA Sells $1.5 Billion in Bitcoin to Fund AI Infrastructure, Signaling a New Era for Miners

MARA Holdings just sold nearly 21,000 bitcoin for roughly $1.5 billion, liquidating 38% of its treasury in a single quarter. The company that once championed HODLing all mined bitcoin "for the foreseeable future" has fundamentally changed course, and the implications extend far beyond one corporate balance sheet.

The move represents the most dramatic strategic pivot by a major publicly traded Bitcoin miner to date, raising a pointed question: are Bitcoin miners becoming something other than Bitcoin miners?

What MARA Actually Did

Between January and March 2026, MARA executed two major bitcoin sales. The larger block, roughly 15,133 BTC sold between March 4 and March 25 for $1.1 billion, went primarily toward repurchasing about $1 billion in zero-coupon convertible notes due 2030 and 2031. The company bought back this debt at a discount, paying approximately $912.8 million in cash to retire $1 billion in face-value obligations.

The strategic centerpiece is a planned $1.5 billion acquisition of Long Ridge Energy & Power in Hannibal, Ohio. The deal includes a 505-megawatt gas-fired power plant and land intended for development into a combined bitcoin mining and AI/high-performance computing campus.

After the sales, MARA's bitcoin holdings dropped from 38,689 BTC to about 35,303 BTC, pushing the company from the second-largest to fourth-largest publicly traded bitcoin holder.

The Board Reversed Its Own Policy

This wasn't a gradual evolution. In March 2026, MARA's board formally reversed its 2024 policy of retaining all mined and purchased bitcoin. The new authorization permits sales of the company's entire treasury (which was 53,822 BTC when the policy changed).

For context, as recently as August 2025, MARA held about 52,477 BTC and sold none that month. The shift from aggressive accumulation to $1.5 billion in liquidations happened in roughly four months.

MARA has indicated this may not be a one-time event. In Q1 2026 earnings communications, management said it expects to fund the remaining Long Ridge consideration through cash on hand, bitcoin-collateralized borrowings, and potentially further bitcoin sales.

The AI Infrastructure Thesis

MARA's internal research projects that global AI power demand could rise 255% by 2030, requiring roughly $5.2 trillion in AI data center investment plus $720 billion for grid upgrades. The company sees itself positioned to capture part of this spending cycle.

The pitch makes a certain logical sense. Bitcoin miners already own power infrastructure, cooling systems, and relationships with utilities. MARA claims roughly 90% of its non-hosted mining capacity could be repurposed for AI workloads. Rather than exclusively mining bitcoin, the company could dynamically allocate capacity between BTC mining and AI compute based on relative profitability.

The broader context supports this opportunity thesis. In 2026, hyperscalers including Alphabet, Amazon, Meta, and Microsoft are deploying hundreds of billions into AI data centers. Nvidia's market value has approached $4.8 trillion, and the company has announced partnerships to build "AI factories" with U.S. national labs and major corporations.

MARA has also expanded internationally, acquiring a 64% stake in French AI infrastructure provider Exaion and positioning for growth in Europe, the Middle East, and Latin America.

How Markets Responded

Equity investors initially rewarded the pivot. Following the March 26, 2026 disclosure, MARA shares rose about 3.6%, and by mid-May the stock was up approximately 25.5% year-to-date.

That said, MARA shares remain down over 23% over the prior 12 months, and Q1 2026 financials showed revenue of $174.6 million (down from $213.9 million in Q1 2025) alongside a net loss of roughly $1.26 billion. The loss was driven largely by a $1 billion unfavorable fair-value adjustment on bitcoin holdings under mark-to-market accounting.

The Bull Case

Supporters argue that MARA is making a rational capital allocation decision. By selling bitcoin above $70,000 and retiring debt at a discount, the company reduces balance-sheet risk and future dilution from convertible note conversions.

The hybrid model could prove valuable. If AI compute becomes more profitable than bitcoin mining during certain periods, MARA can shift capacity. If bitcoin rallies, it can redirect power back to mining. This optionality, bulls argue, is more valuable than simply HODLing bitcoin on a corporate balance sheet.

The $1.5 billion in realized gains also strengthens MARA's position to survive future crypto downturns without forced selling at unfavorable prices.

The Bear Case

Critics raise several concerns. First, selling bitcoin into a rising market sacrifices asymmetric upside. If BTC appreciates significantly from here, MARA gave up potential gains that no AI data center can replicate.

Second, MARA is transitioning from a business where it had genuine competitive advantages (mining expertise, existing hashrate) to one where it faces intense competition from hyperscalers and utilities with far deeper pockets. Google, Amazon, and Microsoft have essentially unlimited capital to deploy on AI infrastructure.

Third, the pivot changes what kind of company MARA is. Investors who bought it as a leveraged bitcoin proxy now own a highly levered infrastructure operator facing execution risk, power market cycles, and a multi-year capex buildout. That's a fundamentally different investment thesis.

Finally, critics warn that miners selling bitcoin into rallies could create headwinds for BTC's price and change how investors think about miners as leveraged proxies. If miners are marginal sellers into strength rather than holders, their stocks may lose some appeal to bitcoin-bullish investors.

What This Means for the Mining Industry

MARA's move may preview a broader trend. Bitcoin miners have long struggled with the economics of their business: significant capex requirements, volatile revenue tied to BTC prices and network difficulty, and energy costs that represent the majority of operating expenses.

AI infrastructure offers potentially steadier cash flows from enterprise contracts and hyperscaler relationships. The power infrastructure that miners have built becomes more valuable if it can serve multiple use cases.

But there's a deeper question here about what bitcoin miners are supposed to be. If the best use of mining infrastructure is something other than mining bitcoin, what does that say about the long-term economics of bitcoin mining itself?

Looking Forward

MARA is betting that the AI infrastructure buildout represents a generational capex cycle worth pivoting toward, even at the cost of liquidating a substantial bitcoin position. Whether this proves prescient or premature likely depends on several factors: how bitcoin performs relative to expectations, whether MARA can execute the Long Ridge acquisition and development effectively, and whether its AI infrastructure can compete with better-capitalized rivals.

For bitcoin-focused investors, the message is worth noting: the era of miners as pure-play bitcoin accumulation vehicles may be ending. The companies that built infrastructure to mine bitcoin are increasingly viewing that infrastructure as valuable for reasons that have nothing to do with bitcoin.

That's not necessarily good or bad. But it is a meaningful change in how major industry participants think about their relationship with Bitcoin itself.