Mining at the Margin: Publicly Traded Bitcoin Miners Are in Survival Mode

The Math Doesn’t Lie: Almost Everyone Is Underwater or Close To It

Here’s the honest arithmetic. After the Bitcoin halving in April 2024 — which cut the block reward from 6.25 BTC to 3.125 BTC — miners needed to do twice the work for the same number of coins. At the same time, the global network hashrate has surged past 900 exahashes per second, meaning competition for those coins has never been fiercer. The result is that the industry’s average all-in cost to mine one Bitcoin, when you include electricity, hardware depreciation, cooling, labor, and corporate overhead, has climbed into the $60,000–$80,000 range for most publicly traded companies.

With Bitcoin sitting at just under $70,000, that’s not a comfortable margin — that’s a crisis in slow motion. Senior analysts at CryptoQuant have pegged the average all-in cost for publicly listed miners at between $60,000 and $80,000 per coin when you factor in corporate overhead and capital expenditures. The best-run operations in the sector are producing Bitcoin for around $55,000 per coin and maintaining a slim but real margin. The worst-run are spending $80,000 or more to produce something worth $70,000 — a business model that only works if you’re willing to drain your treasury and pray the price recovers before the balance sheet gives out.

And unlike a traditional business where you can simply pause production when margins go negative, Bitcoin mining has massive fixed costs. The machines are running whether the price cooperates or not. You can power down inefficient rigs — and many operators are doing exactly that — but you can’t pause the debt service on the billions these companies have raised over the past two years.

Company by Company: Who Can Actually Survive at $70K Bitcoin

At just under $70,000 per Bitcoin, the difference between the efficient operators and the bloated ones is no longer just a performance gap — it’s the difference between staying in business and facing existential pressure. Here’s where each of the major publicly traded miners stands.

MARA Holdings (MARA) is the largest publicly traded Bitcoin miner in the United States, and right now that size is both an asset and a liability. In Q3 2025, MARA disclosed an all-in mining cost of approximately $67,704 per Bitcoin — which, at today’s price of just under $70,000, means the company is producing Bitcoin for essentially what it’s worth on the open market. Margin is measured in hundreds of dollars per coin, not thousands. MARA’s power cost of around $0.04 per kilowatt-hour is genuinely one of the best in the industry, and the company’s 50,000 BTC balance sheet provides a significant buffer. But MARA has also taken on enormous debt — including $1.925 billion borrowed specifically to buy Bitcoin at an average price of $98,529 per coin. Those BTC purchases are now deeply underwater. At $70,000 Bitcoin, that position represents paper losses exceeding $400 million. The company can survive this with its balance sheet, but it’s not a comfortable position for shareholders.

Riot Platforms (RIOT) was one of the highest-cost producers in the sector coming out of late 2024, with all-in costs estimated at roughly $148,000 per Bitcoin during the worst of the post-halving adjustment. The company has been aggressively cutting costs through fleet upgrades and its Texas demand response programs — which effectively have the grid pay Riot to curtail power use during peak periods — but even with that progress, Riot’s current estimated all-in cost per coin remains in the $80,000–$100,000 range by most analyst estimates. At $70,000 Bitcoin, Riot is mining at a loss on a pure production basis. Its cash reserves and Bitcoin holdings provide runway, but the clock is ticking if prices don’t recover. This is a company that needs Bitcoin above $90,000 to breathe easily.

CleanSpark (CLSK) has been one of the more disciplined operators among the large public miners. With 50 exahashes per second of operational hashrate, over 242,000 mining machines, and a focus on low-carbon energy in the American South, CleanSpark has managed to keep its all-in cost per BTC in the $55,000–$65,000 range — which means it still has a thin positive margin at today’s prices. That’s not comfortable, but it’s survivable. CleanSpark holds over 13,000 BTC on its balance sheet and has been expanding into AI and HPC colocation, which could provide meaningful revenue diversification at exactly the moment it’s needed. Of the pure-play miners, CleanSpark is better positioned than most at current Bitcoin prices.

Iris Energy (IREN) is the company that other miners probably envy most right now. Powered almost entirely by renewable energy, IREN has consistently operated as the lowest or near-lowest cost producer among its publicly traded peers, with estimated all-in costs in the $50,000–$60,000 range. At $70,000 Bitcoin, IREN is one of the very few large public miners still generating a meaningful margin on every coin mined. More importantly, IREN has arguably gone furthest in its pivot toward AI data center infrastructure, which provides revenue that is completely independent of where Bitcoin trades. The company’s stock was up 357% year-to-date through late 2025 for a reason — investors can see that IREN has built a business that doesn’t live or die by Bitcoin’s price the way its peers do.

Core Scientific (CORZ) re-listed after its 2022 bankruptcy and has been the most aggressive in pivoting toward AI/HPC hosting. Its landmark 12-year contract with AI hyperscaler CoreWeave was viewed as a game-changer for the sector, and CEO Adam Sullivan has been clear that he prefers the data center business over Bitcoin mining’s volatility. That forward thinking looks prescient right now. Core Scientific’s all-in mining costs are estimated in the $60,000–$70,000 range, meaning its pure Bitcoin mining operations are borderline unprofitable at current prices — but the AI/HPC revenue stream is providing the margin that mining can’t. This is exactly the thesis that justifies the AI pivot, playing out in real time.

TeraWulf (WULF) operates one of the most interesting energy models in the sector, using nuclear and hydroelectric power at its Lake Mariner facility in upstate New York. Clean baseload power at scale is a genuine strategic asset. However, TeraWulf paid approximately $0.081 per kilowatt-hour in Q1 2025, higher than expected, which pushed its costs up sharply. Its estimated all-in mining cost sits in the $60,000–$75,000 range — meaning the company is likely at or below break-even on Bitcoin mining at current prices. The saving grace is its growing AI infrastructure business: the company signed its first AI customer, Core42, for over 70 megawatts of data center capacity, and raised $3.2 billion in senior secured notes to expand. TeraWulf needs to execute on the AI transition faster now that Bitcoin isn’t providing the cushion it once did.

Cipher Mining (CIFR) is a Texas-focused operator that has been quietly executing well. Reported revenue of $72 million in Q3 2025 and a $1.4 billion bond issuance for a Google-linked data center project signal that Cipher has real institutional backing. Its all-in mining costs are estimated in the $55,000–$65,000 range, giving it a narrow but positive margin at today’s Bitcoin price. Like CleanSpark, Cipher is one of the operators that can claim it’s still profitable — barely — on a pure mining basis at $70,000. The AI data center pivot provides important upside if it executes.

Hut 8 (HUT) has transformed itself through its merger with US Bitcoin Corp into an integrated platform spanning self-mining, hosting, and managed infrastructure. Its AI compute leasing revenue is an increasingly important buffer. All-in mining costs estimated in the $60,000–$70,000 range mean Hut 8 is right at the edge of profitability on its mining operations at current prices. The diversification is the reason it can tolerate that.

Bitfarms (BITF) has the most concerning profile at current Bitcoin prices. Despite international diversification and a renewable energy focus across North America and South America, Bitfarms has struggled with execution, and its all-in mining costs are estimated above the industry average. The stock was already trailing its peers in late 2025, and a sustained period of $70,000 Bitcoin would likely force difficult decisions about fleet reduction, asset sales, or additional capital raises under unfavorable terms.

Bit Digital (BTBT) has been aggressively pivoting away from pure Bitcoin mining toward AI and GPU compute services — and that pivot looks like a lifeline right now. Its remaining mining operations carry all-in costs in the $60,000–$70,000 range, but the AI/GPU hosting revenue is increasingly what matters to the investment thesis. Companies like Bit Digital that started diversifying before Bitcoin prices retreated look far smarter today than they did twelve months ago when everyone was focused on hashrate growth.

The AI Lifeline: No Longer Optional

It would be dishonest to write about publicly traded Bitcoin miners right now without stating plainly what the numbers have already made clear: for many of these companies, the AI pivot isn’t a strategic option anymore — it’s a survival mechanism. With Bitcoin under $70,000 and mining costs clustered between $55,000 and $80,000 per coin, the pure mining business model is either barely profitable or actively losing money depending on the operator. The companies that diversified into AI and high-performance computing infrastructure before this price downturn look like geniuses. The ones still primarily dependent on Bitcoin mining revenue are in a difficult spot.

This transition makes structural sense when you understand what these companies actually built. Over the past several years, public miners spent billions constructing something that AI hyperscalers now desperately need: massive power infrastructure, purpose-built data center facilities, and deep operational expertise in managing high-density compute at scale. Bitcoin mining was, in retrospect, the proving ground. When CoreWeave signs a 12-year deal with Core Scientific, or when TeraWulf leases 70+ megawatts to an AI customer, or when Cipher Mining raises $1.4 billion for a Google-linked data center — those aren’t pivots made from desperation. They’re the logical monetization of assets that took years to build, redirected toward a customer base with deeper pockets and more predictable demand than Bitcoin’s price action has ever provided.

The companies managing both streams — IREN, Core Scientific, Cipher, CleanSpark, TeraWulf — have a buffer that pure-play miners simply don’t have at current prices. That buffer is the difference between managing through this downturn and being forced into asset sales, emergency capital raises, or operational shutdowns.

The Leverage Problem: When the Bet Goes Wrong

There’s another issue that needs to be addressed directly, because it’s going to matter enormously if Bitcoin stays at or below $70,000 for an extended period. Over the past two years, many of these companies didn’t just expand mining operations — they borrowed heavily to buy Bitcoin directly as a treasury asset, following a strategy pioneered by Michael Saylor’s Strategy (formerly MicroStrategy).

MARA borrowed $1.925 billion to purchase Bitcoin at an average price of $98,529 per coin. At $70,000 Bitcoin, that position is underwater by more than $400 million on paper. Riot secured $500 million in convertible notes. CleanSpark raised $550 million in new debt. These capital structures made sense when Bitcoin was trading above $100,000 and appeared to have institutional momentum behind it. At $70,000, they represent serious balance sheet risk if prices don’t recover on a timeline that allows the debt to be serviced or refinanced.

This is the piece of the story that tends to get glossed over during bull markets. Bitcoin mining stocks are not simply a leveraged bet on Bitcoin’s price — they’re a leveraged bet on Bitcoin’s price with massive fixed operating costs, aggressive debt loads, and ongoing capital expenditure requirements all layered on top. When the price cooperates, the leverage amplifies gains dramatically. When it doesn’t, the same leverage can accelerate a company’s distress faster than most investors are prepared for.

The Nvidia Ripple Effect: Chips, Circular Financing, and a House of Cards

There’s a downstream consequence to a prolonged mining contraction that hasn’t received nearly enough attention in the financial press — and when you pull on this thread, it connects Bitcoin’s price collapse to one of the most significant and underreported financial structures in the entire technology sector.

The conventional wisdom is that Bitcoin miners don’t really matter to Nvidia because Bitcoin mining runs on ASICs — Application-Specific Integrated Circuits — not Nvidia GPUs. That’s technically true for the actual mining process. But it misses a much larger and considerably more alarming picture.

Here’s where the real connection lives. Over the past two years, as publicly traded miners pivoted aggressively toward AI and high-performance computing infrastructure, they became significant buyers of Nvidia’s H100 and H200 data center GPUs — the most expensive and most coveted chips in the world. Core Scientific, IREN, Cipher Mining, TeraWulf, and others didn’t just announce AI pivots as press release strategy. They went out and actually bought Nvidia hardware to build GPU compute clusters capable of serving AI inference and training workloads. That capital expenditure has been a real and meaningful source of demand feeding into Nvidia’s data center revenue, which now accounts for the overwhelming majority of Nvidia’s business.

When mining economics deteriorate to the point where companies are struggling to fund operations — let alone expansion — the capital spending that would have gone toward Nvidia GPU procurement gets deferred, reduced, or cancelled entirely. A Riot Platforms or Bitfarms that is burning cash trying to survive at $70,000 Bitcoin is not simultaneously writing checks for H100 clusters.

But that’s just the surface layer. What lies beneath is far more consequential.

Nvidia Has Essentially Become a Bank — And That’s a Problem

Over the past two years, something extraordinary has happened in the AI hardware market that has received far too little mainstream scrutiny: Nvidia has quietly transformed from a chip company into what is functionally a financial institution. And the mining sector’s distress could be one of the sparks that exposes it.

Here’s how the machine works. Companies that want to build AI compute infrastructure — including the former Bitcoin miners pivoting toward AI hosting — need enormous quantities of Nvidia GPUs. H100 and H200 chips cost anywhere from $30,000 to $40,000 each, and building meaningful AI infrastructure requires thousands of them. Most companies don’t have the cash to buy that hardware outright. So a new financial structure emerged: GPU-backed debt, where Nvidia chips are used as collateral for loans from private credit firms.

The numbers are staggering. CoreWeave — which began life as a crypto miner called Atlantic Crypto before pivoting to AI cloud services — has built its entire business on this model, amassing over $12 billion in GPU-collateralized debt by 2025. That includes a $2.3 billion GPU-backed facility in 2023, a $7.5 billion private credit facility in 2024, and a $2.6 billion term loan in 2025, all to fund Nvidia hardware purchases. Lambda secured a $500 million GPU-backed loan and entered a $1.5 billion GPU leasing deal. The broader “neocloud” sector — AI infrastructure companies that are effectively the new miners of the digital economy — now carries more than $20 billion in GPU-backed debt, creating what analysts are calling an entirely new private-credit asset class. Blackstone, PIMCO, Carlyle, and BlackRock are all active lenders in this market.

Now here’s where it gets circular in a way that should concern anyone paying attention. Nvidia didn’t just watch this ecosystem develop from the sidelines. It actively participated in and enabled it. Nvidia has taken equity stakes in CoreWeave, OpenAI, and other key customers. It committed to purchasing up to $6.3 billion in unsold CoreWeave cloud capacity as a backstop — meaning Nvidia is guaranteeing demand for a customer that uses that guarantee to take on more debt to buy more Nvidia chips. It committed up to $100 billion to OpenAI, which will reportedly lease rather than purchase Nvidia chips, creating a circular flow where Nvidia’s investment comes back to Nvidia as revenue. In September 2025, Nvidia also announced it would lease 18,000 GPUs from Lambda for $1.5 billion — a company that had used GPU-backed debt to finance those very chips in the first place.

Short seller Jim Chanos summarized the dynamic bluntly in late 2025: the AI infrastructure buildout is real, the customers are real, but the capital structure holding it together is increasingly debt-financed against hardware that is depreciating rapidly. H100 GPU rental rates, which peaked around $8 per hour in 2023 during the shortage, had already collapsed to below $2 per hour by late 2024 as supply expanded — a 75% decline in the revenue-generating capacity of the very assets being used as loan collateral.

How Miner Distress Feeds Into This System

The connection back to Bitcoin mining is this: the former miners who pivoted to AI — Core Scientific, IREN, Cipher, TeraWulf — are participants in this same GPU-backed financing ecosystem. They bought Nvidia hardware, often using debt, to stand up AI compute services. If Bitcoin prices stay depressed and mining revenue dries up, these companies face a cascading problem: mining cash flows aren’t available to service debt obligations, the AI hosting revenue may not yet be sufficient to cover the gap, and the GPU assets underpinning their balance sheets are depreciating at the same time they need to be refinanced.

CoreWeave alone faces $986 million in debt payments due in 2025 and $4.2 billion in 2026 — all of it backed by GPUs whose collateral value is shrinking as newer Nvidia chips come to market and rental rates continue to compress. The math is uncomfortable: loans secured by H100s that are worth $40,000 today may be secured by assets worth $20,000 or less by the time those loans mature, because Nvidia’s next-generation Blackwell GPUs are already in deployment and the H100’s premium is eroding.

When distressed mining companies are forced to liquidate GPU assets to raise liquidity, those used H100s and H200s flood secondary markets at steep discounts. That softens demand for new chips and gives potential buyers a cheaper alternative, directly impacting Nvidia’s ability to sustain current pricing power. More significantly, it undermines the collateral value underpinning the entire $20 billion GPU-backed debt market — creating the potential for a synchronized tightening of credit availability across the entire AI infrastructure sector at exactly the moment these companies are facing their heaviest debt repayment schedules.

The scale of Nvidia’s direct exposure shouldn’t be overstated — its revenue is ultimately driven by Microsoft, Google, Amazon, Meta, and the foundational hyperscalers who operate at an entirely different financial scale. But the circular financing structure Nvidia has built with its second-tier customers — the very same companies that were once Bitcoin miners — represents a genuine systemic risk that the market has been slow to price. If enough of these leveraged AI infrastructure plays hit the wall at the same time, the resulting forced GPU liquidations and credit tightening could create a headwind for Nvidia’s data center revenue that most analysts haven’t modeled.

The irony is almost poetic. Bitcoin miners borrowed billions to buy Nvidia hardware, intending to escape Bitcoin’s volatility by building AI infrastructure. The AI infrastructure they built is financed with GPU-backed debt that depreciates as fast as Bitcoin can fall. And Nvidia, the company that sold them all those chips, has made itself a creditor, equity holder, and revenue backstop for the ecosystem holding the whole thing together. At $70,000 Bitcoin, we may be watching the first stress test of how sturdy that structure actually is.

Looking Forward: What Needs to Happen

The path forward for publicly traded Bitcoin miners is not complicated to understand, even if it’s difficult to execute. The efficient operators — those with sub-$60,000 all-in mining costs — can survive the current environment and potentially emerge stronger as less efficient competitors are forced to power down equipment or exit the market entirely. A reduction in active mining operations would lower network difficulty over time, improving margins for those who remain. This is how Bitcoin mining has always self-corrected through difficult periods.

The companies with meaningful AI/HPC revenue streams have a second lever entirely independent of Bitcoin’s price. That revenue pays the bills while the balance sheet’s Bitcoin holdings either recover value or are managed down carefully. For these operators, a $70,000 Bitcoin is painful but manageable.

The companies that are high-cost pure-play miners with significant debt loads are in genuinely precarious territory. A sustained period at current prices — particularly if it stretches into mid-2026 — would likely force some combination of fleet reduction, asset sales, equity raises at unfavorable valuations, or in the worst cases, restructuring conversations with creditors.

What’s clear is that this moment is the reckoning the industry was warned about when half the sector was borrowing billions at peak enthusiasm. The miners who survive — and eventually thrive — will be the ones who spent the last two years building something that doesn’t require Bitcoin to trade at $100,000 to justify their existence. At just under $70,000 per coin, with costs nearly matching that price for most operators, we’re about to find out who actually did the work.

This article is for informational and journalistic purposes only and does not constitute investment advice. Bitcoin mining stocks are highly volatile and speculative investments. Always consult a qualified financial advisor before making investment decisions.