The chart is a symptom, not the cause. Bitcoin’s hashrate dropped 20.6% from its all-time high in May 2025, settling at 914 EH/s by late August. Meanwhile, the price climbed 34.9% over the same period. This is the second time in Bitcoin’s history that price and hashrate have diverged so sharply. The first was in 2012, during the post-Mt. Gox recovery. Back then, the divergence lasted only weeks before miners returned. This time, the gap is widening. And the cause isn’t a technical flaw in Bitcoin’s consensus—it’s a structural reallocation of resources toward AI. Code doesn’t lie. The block timestamps confirm the network is running smoothly, averaging 9 minutes and 56 seconds per block. Difficulty adjusted downward multiple times, yet hashrate refuses to follow the historical script. Something fundamental has changed.
Miners are not leaving because of a bug or a regulatory crackdown. They are leaving because the opportunity cost of mining Bitcoin has become too high. The AI boom has created a parallel market for high-performance computing, and miners—who already own the power contracts, the real estate, and the cooling infrastructure—are uniquely positioned to serve it. Riot Platforms signed a 20-year colocation agreement with Anthropic. IREN cut its Bitcoin mining capacity by 30% and redirected power to its AI cloud service. TeraWulf is now generating more revenue from HPC than from Bitcoin. These are not anecdotal moves. They are strategic pivots backed by shareholder capital.
Core: The Data Behind the Exodus
Let’s start with the numbers. The seven-day average hashrate peaked at 1,175 EH/s in May 2025. By August 25, it was 914 EH/s—a decline of 261 EH/s. That’s roughly 2.6 million Antminer S19s going offline, assuming a 100 TH/s per unit. But the interesting part is that the network difficulty adjusted down by 14% over the same period, from 95.2 trillion to 82.1 trillion. In a normal cycle, this would have made mining more profitable, pulling hashrate back. But the Puell Multiple, a measure of miner revenue relative to its one-year moving average, is at 0.73—the 16th percentile. Miners are still earning less than their historical average, even after the difficulty drop.
Hashprice, the dollar-denominated revenue per PH/s per day, did improve from a low of $32.50 in June to $39.36 by August. That’s a 21% recovery. But it’s still below the $50 threshold that most industrial miners consider break-even. The hashprice improvement is a direct result of the price increase and the difficulty adjustment. Yet the hashrate didn’t bounce back. Why? Because the miners who left have committed their power to AI contracts with guaranteed revenue streams. They are no longer marginal participants who can toggle back on when Bitcoin becomes more profitable. Signal over noise. Always. The noise is the hashprice recovery. The signal is the permanent lock-in of power capacity.
The Lock-In Effect
I’ve been watching this market since 2017, when I reverse-engineered the 0x protocol’s smart contracts to find a re-entrancy bug. That experience taught me that code is the ultimate truth. The same applies here. The code is Bitcoin’s difficulty adjustment algorithm (DAA), which responds to hashrate changes with a two-week lag. The DAA does its job. But it cannot reverse a decision backed by a 20-year contract. Riot’s deal with Anthropic locks in 200 megawatts of power for the next two decades. That’s roughly 20 EH/s of potential Bitcoin hashrate that will never return, regardless of how high Bitcoin’s price goes.
IREN, once a pure Bitcoin miner, now derives 40% of its revenue from AI cloud services. Its CEO stated publicly that the company will not deploy new Bitcoin mining rigs in 2025. Instead, it’s ordering Nvidia H100 GPUs for AI inference. The capital expenditure is being redirected. This is not a cyclical downturn. It’s a structural shift in the allocation of physical resources. The chart is a symptom, not the cause. The cause is the superior economics of AI compute.
Let’s quantify this. A typical Bitcoin mining site with 100 MW of power can generate roughly $10 million per month in revenue at current hashprice, assuming 10 EH/s. The same power, repurposed for AI inference, can generate $30-40 million per month, depending on the GPU density and customer contracts. The margin is also higher because AI workloads are less volatile and require less capital expenditure on ASICs. The decision is obvious for any rational operator.
Contrarian: The Market Is Misreading the Risk
Mainstream analysis frames this as a temporary miner capitulation. The narrative goes: “Bitcoin price will recover, difficulty will drop further, and miners will return.” I’ve heard this before—during the 2018 bear market, the 2020 halving, and the 2022 LUNA crash. In each case, the cycle repeated. But the cycle is broken. The reason is that the “miner surrender” mechanism relies on a homogeneous group of operators who all face the same cost structure. That assumption no longer holds. Miners are now bifurcated: those who went all-in on AI (IREN, TeraWulf, Hive Digital) and those who are doubling down on Bitcoin (MARA, Riot, Bitdeer). The latter group is still adding hashrate, but their growth is being offset by the former group’s subtraction.
There’s another blind spot. The market assumes that hashrate is a lagging indicator of price, but it’s also a leading indicator of security. Bitcoin’s market cap hovers around $1.2 trillion. The cost to execute a 51% attack is roughly $10 billion in hashrate rental, assuming the attacker controls 50% of the network for 12 hours. If hashrate drops another 20%, that cost falls to $8 billion. That’s still prohibitively high, but the direction is concerning. The security premium that Bitcoin commands over other cryptocurrencies is directly tied to its hashrate. If that premium erodes, the valuation argument weakens.
Back in 2022, during the LUNA/UST crash, I published a minute-by-minute forensic timeline of the algorithmic failure. I saw then how a seemingly robust mechanism—the LUNA-UST burn/mint model—could collapse when the arbitrageurs stopped trusting the system. The same psychological threshold exists for Bitcoin. If enough market participants believe that hashrate will never recover, the safe-haven narrative takes a hit. The chart is a symptom, not the cause. The cause is the shifting perception of Bitcoin’s security.
Takeaway: The Next Watch Points
Sleep is for those who can’t trade. The next 90 days will determine whether this is a structural realignment or a temporary deviation. I’m watching three things. First, the hashrate recovery after the next difficulty adjustment. If it fails to climb above 950 EH/s within two weeks, the lock-in effect is real. Second, the AI cloud revenue reported by IREN and TeraWulf in their Q3 earnings. If they show sustained growth, expect more miners to pivot. Third, the Bitcoin price. If it stays above $65,000, the opportunity cost of not mining will eventually draw some hashrate back. But if it dips below $50,000, the spiral could accelerate.
The contrarian take is that this is actually bullish for Bitcoin’s long-term price. Fewer miners means less daily sell pressure from block rewards. With the halving already reducing new supply from 900 to 450 BTC per day, a further drop in active miners could cut sell pressure to 300 BTC per day. That’s a 66% reduction from pre-halving levels. If demand stays constant, price must rise. But this argument assumes that the remaining miners are profitable and hodl. The data so far shows that miners are selling more of their production than they did in 2024. The correlation between hashrate and sell pressure is not linear.
Code doesn’t lie. The Bitcoin network’s DAA will continue to adjust, but it cannot force a miner to turn on a machine that is locked into a more profitable contract. The market is pricing this as a temporary dip. I see it as a permanent transition. The question is: will Bitcoin find a new equilibrium at a lower hashrate, or will the security premium collapse? I’ve been in this industry for 20 years. I’ve audited protocols, tracked wallets, and analyzed market microstructures. This is the most important structural change since the 2017 ICO boom. Stay alert. The signal is already there. The noise is just beginning to fade.
