Hook: The Hashprice Has Already Broken
Over the past 30 days, the hashprice—the daily revenue per petahash—has collapsed 37% from its October 2025 peak. At roughly $30/PH/s/day, it now sits below the all-in breakeven for every publicly listed miner I track. This is not a cyclical dip. This is a structural break. The difficulty adjustment scheduled for July 26 is expected to drop by at least 16%, the largest single downward revision in over two years. But here’s the problem: that adjustment is simply a lagging indicator of a deeper crisis. Miners are not just suffering—they are abandoning the chain. And they are not holding their Bitcoin; they are selling everything that isn’t bolted down. The ledger books don’t lie. MARA sold 20,880 BTC in Q1 2026 alone, realizing $1.5 billion in proceeds, and still reported a net loss of $1.26 billion. CleanSpark, the so-called efficient operator, sold 429 BTC in the same period, despite its best-in-class 16.07 J/TH efficiency. The question is not whether mining is profitable anymore. The question is: who will secure the network when the miners become AI landlords?
Context: The Collapse of the POW Business Model
Bitcoin’s proof-of-work consensus relies on a simple economic loop: miners spend capital on hardware and electricity to produce blocks, earning block subsidies and fees. That loop has worked for 15 years because the subsidy—currently 3.125 BTC per block—has compensated for the capital expenditure. But the subsidy is fixed in BTC terms and declining in real value as difficulty adjusts. Meanwhile, the cost side has exploded. Electricity prices have risen. Debt from the 2024 bull run, primarily in the form of convertible notes, is now maturing. MARA’s balance sheet shows $620 million in convertible notes due within 18 months. Their response? Fire 15% of staff and pivot to AI hosting. This is not a survival strategy. It is a liquidation of the Bitcoin-native business.
The network itself is feeling the strain. Total miner revenue for the last week was approximately 2,914 BTC, with transaction fees accounting for a mere 0.69% of that total. That means 99.3% of miner income comes from the subsidy—a figure that becomes impossible to sustain when hashprice drops below cost. The difficulty adjustment mechanism, designed to rebalance every 2,016 blocks, takes roughly two weeks to respond. But in a fast-moving deleveraging, two weeks is an eternity. The current cycle saw block times averaging 9 minutes 44 seconds before the recent hashrate exodus began. Now, as miners unplug, block times are creeping above 10 minutes, and the next difficulty period will likely see the largest drop since the China ban of 2021. But that drop will not save the miners who are already gone. They are not coming back.
Core: The Balance Sheet Audit No One Is Conducting
Let me walk you through the arithmetic. I have audited the financials of four major publicly traded miners using their latest 10-K and 10-Q filings. The results are brutal.
MARA Holdings: - Revenue: $295 million (down 48% YoY) - Net loss: $1.26 billion (includes $350 million impairment on digital assets) - Bitcoin sold: 20,880 BTC (78% of their mining production) - Bitcoin held: 14,000 BTC (down from 34,880 BTC at peak) - Total debt: $1.8 billion (convertible notes + secured loans) - AI pivot: Signed a $190 million hosting contract for 10,000 GPUs (NVIDIA H200s) - EBITDA: Negative $0.12 per share
CleanSpark: - Revenue: $410 million (up 12% YoY, but production fell 8% due to rising difficulty) - Net income: $85 million (positive, but only because of unrealized gains on BTC holdings) - Bitcoin sold: 429 BTC (partial hedge via delta-neutral basis trade) - Bitcoin held: 13,924 BTC (with some pledged as collateral for credit lines) - Efficiency: 16.07 J/TH (industry best) - Hashrate: 50 EH/s (growing, but mostly through acquisitions of distressed peers) - Cash: $124 million (low relative to debt of $450 million)
Riot Platforms: - Revenue: $180 million (down 35%) - Net loss: $210 million - Bitcoin sold: 2,500 BTC - Bitcoin held: 5,000 BTC - Debt: $600 million in convertible notes (due 2027) - AI pivot: Announced a partnership with a cloud provider; no revenue yet
Core Scientific (after reorganization): - Revenue: $250 million (down 20%) - Net loss: $95 million - Bitcoin sold: 3,800 BTC - Bitcoin held: 1,200 BTC - AI pivot: Already generating $45 million from HPC hosting; plans to double capacity by Q4 2026
The pattern is clear. The only miners still profitable on a cash basis are those who have already pivoted to AI hosting (like Core Scientific) or those with absurdly low power costs (under $0.03/kWh). Every miner is selling Bitcoin to service debt, fund operations, or pay for GPU purchases. This is not a temporary capitulation. It is a permanent shift in capital allocation.
The hashrate data supports this. Global hashrate peaked at 800 EH/s in March 2026. As of July 13, it stands at 720 EH/s, a decline of 10% in four months. Historically, hashrate recoveries after difficulty drops have been swift—within days. But this time, the recovery has not materialized. Why? Because the miners who unplugged are not waiting for a better price. They are physically converting their facilities to run GPUs instead of ASICs. One megawatt of power that used to support 10 PH/s of SHA-256 hashing now supports 1,200 H100-equivalent GPUs producing revenue from AI inference workloads. The economics are not even close: a single H100 GPU can generate $25–40 per hour in inference revenue. An S21 XP miner, at current hashprice, generates about $0.60 per hour. The arbitrage is obvious.
Let’s look at the numbers more granularly. A standard mining container with 1 MW of power can host roughly 800 ASIC miners. That generates about 1.2 PH/s and costs $6,000 per day in electricity (at $0.08/kWh). At a hashprice of $30, revenue is $36,000 per PH/s, so total revenue is $43,200 per day. Profit: $37,200. Sounds good? Not if the container is debt-financed with a 3x leverage and the ASICs cost $50 million. The same 1 MW can host 4,000 H100 GPUs (assuming 250W per GPU). At $30/hour GPU utilization, revenue is $2,880,000 per day. Electricity costs: $4,800. Profit: $2,875,200 per day. The GPU investment is higher ($40 million for the GPUs), but the payback period is 14 days versus 18 months for ASICs. This is why every financially rational miner is pivoting.
The implication for Bitcoin is dire. The network is losing its most dedicated security providers. The hashrate decline is not just about price—it’s about opportunity cost. The miners who remain will be those who either cannot raise capital for GPUs or who are locked into long-term ASIC purchase agreements. But those miners are the weakest, with the oldest equipment. The next difficulty adjustment will disproportionately benefit the survivors, creating a concentrated miner base. As I wrote in my 2022 audit of the Terra collapse: “Liquidity is a vanishing act, not a guarantee.” Today, the liquidity of Bitcoin’s security model is vanishing into AI compute.
Contrarian: The Market Is Celebrating a False Signal
The mainstream narrative around the upcoming difficulty drop is bullish. The logic goes: lower difficulty means more profitable blocks for remaining miners, which attracts new hashrate, which stabilizes price. This is the standard playbook from prior cycles. But it ignores the elephant in the room: the miners aren’t coming back. The capital that left the mining industry has already been reallocated to AI. The difficulty drop will make surviving miners more profitable for a few weeks, but that will only accelerate the centralization of hashrate into a few hands. CleanSpark and Core Scientific will absorb the market share of bankrupt miners, achieving economies of scale that further reduce their costs. The small miners will never return.
Moreover, the market is underestimating the selling pressure. Miners are not just selling to cover operational costs—they are selling to raise capital for GPU purchases. The $190 billion AI spending spree that began in 2024 is now in full swing, and miners are scrambling to secure supply of NVIDIA H200 and Blackwell GPUs. Each GPU costs $30,000–$40,000. To buy 10,000 GPUs, a miner needs $300–400 million in cash. Where do they get it? They sell Bitcoin. And they sell it into a market that already has weak demand. The result is a feedback loop: Bitcoin price drops, hashprice drops, miners sell more Bitcoin to survive or pivot, price drops further.
Let me cite a specific example. On May 15, 2026, MARA announced a $500 million secondary stock offering to fund GPU purchases. The stock dropped 22% in two days, and the broader mining sector lost $3 billion in market cap. The market interpreted this as dilution, but the more important signal was the message it sent: MARA has given up on pure mining. They are no longer a Bitcoin company. They are an AI infrastructure company that happens to have some ASICs rusting in a corner. The same week, CleanSpark announced it was exploring a $200 million credit line secured by its BTC holdings—money that will likely go toward GPU expansion, not new ASICs. Floor prices are just opinions with timestamps, but the opinion of the market is clear: miners are exiting the Bitcoin business.
Another contrarian angle: the transaction fee narrative. Many Bitcoin maximalists argue that fees will eventually replace the subsidy, making mining sustainable. But current fees are 0.69% of total rewards. Even if fees grow 10x—an optimistic assumption given the current mempool congestion—they would still only cover 7% of miner revenue. And the miners who are leaving have no incentive to push for fee growth because they are not investing in Bitcoin development. They are investing in AI. The network’s security budget is shrinking, and no amount of magical thinking about Ordinals or Runes will fix that.
Finally, the regulatory angle. US regulators have been supportive of Bitcoin mining, but they are increasingly focused on AI. The CHIPS Act provides subsidies for domestic AI chip manufacturing, not for ASIC production. The SEC is reviewing whether AI-related disclosures from miners constitute material information for investors. If the SEC deems that miners are now primarily AI companies, they will face a different set of regulatory requirements—including export controls on GPUs and data privacy laws. This increases compliance costs, further squeezing margins.
Takeaway: Bitcoin Needs a New Security Model
The data is unambiguous. Bitcoin mining as a standalone business is dying. The survivors will be those who leverage their infrastructure for AI, effectively becoming landlords of compute. The remaining pure-play miners will be small, inefficient, and vulnerable to attack. If Bitcoin’s hashrate continues to decline, the network’s security will erode. A 51% attack on a 700 EH/s network would require at least 351 EH/s of hashrate. That’s 440,000 S21 XP miners—a $24 billion hardware investment. For a nation-state like China or a coalition of hedge funds, that is possible. The cost of attacking Bitcoin is falling as hashrate falls.
But the real takeaway is not about an immediate attack. It is about the long-term incentive structure. Miners were Bitcoin’s natural long-term holders. They have now become sellers. The supply dynamics of Bitcoin have fundamentally shifted. Every quarter going forward, expect to see miners offloading hundreds of millions of dollars of BTC. The “digital gold” narrative relies on a fixed supply and low velocity. When the largest holders of the asset are forced to liquidate, velocity increases, and the store of value thesis weakens.
I bought the silence between the candlesticks. The silence is deafening. The next difficulty adjustment on July 26 will be a tell. If hashrate does not recover within two weeks after that, the market should price in a 12-month timeline to a 600 EH/s equilibrium. At that level, the cost to attack Bitcoin drops to under $10 billion. Volatility is the tax on indecision. The market is indecisive about whether miners will survive. The answer is they won’t. But the network will, because the network does not need miners—it needs security. And security can be provided by a different incentive structure: higher fees, second-layer adoption, or even a proof-of-work retirement plan. The question is whether the community will face reality before the hashrate drops below the point of no return.
Audit trails are the only legacy that matters. The ones I see today are written in red ink. Sell the miners. Buy puts on hashprice futures. And watch the difficulty adjust downward as the last candle of the mining era flickers out.