There is a sound a dying mining machine does not make. When Bitcoin difficulty fell another 10.09 percent this June โ the third-deepest descent of the ASIC era โ the hashrate charts behaved like a coastline at low tide: the water retreats quickly, and the shore left behind looks less like nature and more like architecture waiting for a new tenant. The network has gone 316 days without a single fresh all-time high in computing power. From the 1.1 zettahash peak of October 2025, the network has settled near 900 exahashes, a withdrawal of roughly 20 percent that the word "correction" fails to contain. Hashprice โ that merciless number telling a miner what one petahash earns in a day โ now hovers around $31.8, after touching $23.9 in the first quarter, its weakest reading since 2018. An industry that once measured itself in zettahashes now measures itself in margins thinner than prayers.
This is what a migration looks like when it happens in the physical world. It does not announce itself with a headline. It simply recalibrates, 2,016 blocks at a time, quietly, while the rest of the market stares at a Bitcoin price that has not decided whether $74,000 is a floor or a warning.
To understand what drained the money out of the room, we have to stop talking about hardware and start talking about what hardware was always standing on: an electricity contract. In the years before 2025, the public miner was the purest expression of the Bitcoin leverage trade โ a balance sheet that converted cheap debt into even cheaper kilowatt-hours, and kilowatt-hours into coins. It was a beautiful machine while it lasted. But the equation broke at both ends at once. Bitcoin fell from $125,000 to a brutal, sobering range near $77,000 to $80,000, just as an AI build-out of historic proportions began bidding for the same substations, the same transformer lead times, the same flat, cool, arid land where data centers have always gone to hide from the sun and compute in peace.
I have watched this industry from before the ASIC era, through the ICO noise, through DeFi Summer, through the institutions arriving and the institutions pretending they had always been there. I have read every hashprice table since they became a form of literature. And I can tell you what so many headlines miss: What we are witnessing is not the death of mining. It is the dismantling of mining's old identity and the auctioning of its parts. The machines did not fail. They were rerouted.
The First Separation: An ASIC Cannot Dream
The most important technical fact in this entire transition is almost too obvious to repeat: a Bitcoin mining machine and an AI compute cluster are not interchangeable, and pretending they are has produced an entire genre of confused commentary. An Antminer S21 executes SHA-256 hashing and nothing else. It has no memory bus for large models, no tensor core, no vocabulary beyond the one word it was born to say. It is a hammer that believes the entire universe is a nail. When Core Scientific pivots its sites toward high-performance computing, it is not flipping a switch inside the machine; it is ripping out the machine and building a different building around the same electrical feed. This is not a technical upgrade. It is a change of religion.
That is why the phrase "AI is replacing miners" is dangerously imprecise. What is actually being replaced is the priority claim on power. AI operators do not look at a mining facility and covet its racks. They covet what the racks are plugged into. Analysts who study these deals โ and I have now watched a full cycle of them โ will tell you that an AI tenant evaluates a site by its substation capacity, its grid interconnect, its optical fiber, and its ability to reach 100 megawatts before the next rainy season. The mining hardware, the ASICs that once defined the business, are the part most easily thrown away. The power contract is the soul. The rest is furniture.
This explains the strange arithmetic we saw all through 2025 and into 2026. Core Scientific reported a self-mining gross margin of negative 56 percent โ meaning every coin it mined itself cost more capital than it returned โ while its data center leasing business generated nearly $80 million in gross margin from the same dirt. A company losing money on Bitcoin while making money on the building that used to mine it is not a company in decline. It is a company in metamorphosis. TeraWulf reached a more dramatic milestone: HPC lease revenue now accounts for roughly 71 percent of its income, and its counterparty is backed by Google. The miners who hedged their destiny are not the ones in pain. They are the ones writing the new rulebook.
None of this is risk-free, and it would be dishonest to write as though a lease agreement is a form of salvation. Core Scientific financed its transformation with $3.3 billion of high-yield debt in April, and it now faces a shareholder lawsuit accusing management of exaggerating its readiness for the AI workload it promised to serve. Its acquisition by CoreWeave โ a $9 billion deal that made a mining company into a footnote inside a cloud giant's real estate portfolio โ tells you everything about who holds the power in this relationship. The miner does not rent to the AI cloud. The AI cloud absorbs the miner.
Yet even in that absorption, there is an honesty that pure mining had lost. The old business model asked a volatile asset price to pay for fixed infrastructure costs, and for years the scale of the lie was hidden by a bull market. The bear market did not create the problem. It merely refused to keep funding the fantasy.
The Thirty-Two Thousand Coin Vigil
The most revealing number in the entire transition is not hashrate or difficulty. It is the public miners' balance sheet. In the first quarter of 2026, listed mining companies sold 32,000 Bitcoin โ a single-quarter record that exceeded the total they sold in all of 2025. The old narrative said miners were Bitcoin's most faithful accumulator class, the natural buyer who never sells because the machine's entire purpose is to hold what it mints. That narrative was always a bull-market convenience. In a bear market, when the transformation demands capital, the accumulation stops, and the coins move.
I have been present for enough cycles to know what this kind of selling feels like from inside a community. It feels like betrayal to those who believed the miner would be the last diamond hand. But let us be precise about what is happening. The miner who sells Bitcoin to finance an AI data center is not gambling on a higher coin price later. He is diversifying out of the dependence on that coin price entirely. The awkward beauty of this is that mining companies may someday stop being forced sellers precisely because they have stopped needing to be miners. Once their electricity is under long-term AI lease, the pressure to liquidate treasury reserves vanishes. The 32,000-coin capitulation of Q1 2026 may be remembered as the fever breaking, not the disease worsening.
There is a kind of spiritual weight in watching a miner part with its own production. Mining Bitcoin has always had a strange purity: the machine does the work, the coin arrives, and the human must choose whether to hold or to sell. To own nothing is to feel everything, deeply. To sell everything you mine is to become a mercenary rather than a sovereign. But mercenaries, at least, are honest about their fee structure. And a miner that sells at $74,000 to pay for the infrastructure that will outlive the bear market may be more honorable than a miner that quietly defaults on its obligations while whispering promises of a future 100x.
The Second Separation: The Grid Discovers a Flexible Tenant
The next phase of this story will be determined not by chipmakers in Taiwan but by grid operators and renewable energy developers. There is a simple structural fact that AI and Bitcoin are converging on: both are hungrier than the grid is elastic. But their hunger is different in kind, not just in degree. AI data centers require their power to be constant, dense, and predictable, because an interrupted training run is a destroyed investment. Bitcoin miners do not. An ASIC can be switched on and off almost casually, treating electricity as a waveform to ride rather than a resource to hoard.
That single difference creates the market separation the original article's conclusion gestured toward. Instead of AI crowding out Bitcoin from a finite pool of power, the two are being sorted, tenant by tenant, into different classes of the same infrastructure. The premium, always-on electric load goes to AI. The residual, intermittent, sometimes-nowhere-else-to-go excess goes to the machines that can tolerate the grid's tantrums.
This is precisely what the French energy giant ENGIE is studying in Brazil. The company is evaluating the placement of Bitcoin mining directly at its solar projects, using machines as a kind of electrical sponge for energy that the grid cannot absorb at certain hours. If a solar farm is curtailed because the grid simply has nowhere to send the electrons, and a mining container sitting on the same site can convert those otherwise-wasted electrons into hashes, then mining is no longer competing with AI for power. It is competing with nothing. It is occupying a space that was otherwise pure loss.
Brazil has made its intentions unambiguous by eliminating import tariffs on high-efficiency mining hardware through January 2028. The message is not "we love Bitcoin." The message is "we have an energy absorption problem, and we have noticed that Bitcoin is the most patient buyer of last resort the modern grid has ever encountered." In this framing, mining becomes something the energy industry calls a flexible load โ and this is the language that matters. Not "proof of work." Not "digital gold." A flexible load is the term of engineers who have studied the problem and whose spreadsheets say the machine is useful even before it mints a single coin.
Now I must be careful with my own enthusiasm, because I have seen idealistic frameworks fail when they met a physical constraint. The curtailment-inspired mining model will not scale beyond regions where renewable generation overbuilds the transmission capacity around it. It will not operate where the grid is always full and always profitable. But the model does not need to power the entire network. It only needs to become the marginal source โ and the marginal source, in a competitive industry, is exactly what determines the price everyone else receives.
The Sovereignty Ledger: What 900 Exahashes Still Means
Amid all the talk of retreat, we should not lose sight of what remains standing. A network at 900 exahashes is not a weakened network. It is a network that has given back a fraction of its exuberance while still resting at a level that would have seemed like science fiction five years ago. The cost of mounting a 51 percent attack against hundreds of exahashes remains in the tens of billions of dollars, before considering that any successful attack would collapse the value of the very asset the attacker hoped to control. The security assumption of Bitcoin was never that hashrate would always rise. It was that difficulty would always adjust, and that the honest majority of economic actors would always have more to gain from preserving the system than from breaking it. Trust is not a transaction; it is a resonance. And resonance does not disappear when some of the singers leave the choir. It simply lowers into a register the remaining voices can support.
The difficulty adjustment mechanism โ running every 2,016 blocks, unaware and uncaring of corporate strategy memos โ is the quiet guardian of the entire transition. When miners switched their machines off, the network did not panic. It recalculated the price of admission until the survivors found themselves profitable again. The third-largest difficulty drop of the ASIC era is not evidence of sickness. It is evidence of equilibrium doing its ancient work. What other financial system can lose one-fifth of its total productive capacity and then simply lower the bar for everyone still standing? The algorithm asks a question that no central banker can ask: how many of you are willing to keep working at this margin? And it adjusts the world until the answer is just enough.
Let me also anchor this in something I know from direct experience rather than from chart-watching. During the worst weeks of 2026's first quarter, when hashprice brushed $23.9 and the obituaries for mining began appearing in mainstream financial media, I spent hours with younger miners who had entered the industry during the euphoria of 2024 and never seen an adverse difficulty cycle. Their fear was not about bankruptcy โ it was about identity. They had been told that mining was the truest form of Bitcoin belief, and now the machines in their warehouses were worth more as real estate than as mining equipment. I understood the vertigo. I felt something close to it in 2018 during my silent audit of the charity token that looked, on paper, like a use case that would justify a bull market, and turned out in practice to be a structure that could drain itself of $2.5 million through a reentrancy loop. What I learned then was that the thing you believe in is often not the thing that survives. What survives is the underlying principle, reshaped, often humiliated, sometimes barely recognizable.
What reshaping awaits the mining industry is visible in its own business models. The numbers are unforgiving. A modern-generation machine, an S21 XP at 13.5 joules per terahash or an S23 Hydro at 9.5 joules per terahash, can survive electricity rates up to $0.088 to $0.124 per kilowatt-hour at current hashprice. The aging S19 class, still operating at 25 joules per terahash or worse, requires power below $0.055 โ electricity that barely exists in open markets outside of the most desperate energy-surplus zones. This is not a market failure. This is the natural selection that every industrial revolution imposes on its predecessors. The new machines are roughly seven times more efficient than the hardware they displaced. Mining is not dying. It is being distilled.
The Contrarian Read: AI Might Decentralize Bitcoin More Than Bull Markets Did
Here is the conclusion I did not expect to reach. The conventional intellectual laziness says AI is the centralizing force โ the giant data center empires swallowing an independent mining industry into the jaws of hyperscale clouds. The data tells another story. Public listed mining companies controlled a shrinking share of the network's hashrate, falling 13.4 percent from the fourth quarter of 2025 through the second quarter of 2026, precisely because their capital is now fleeing toward AI leases. The hashrate that remains โ and the growth that will eventually return โ is increasingly coming from private operators, energy producers, and jurisdictions like Brazil, Paraguay, Ethiopia, and Oman, where a Bitcoin mine is not a public stock but a physical facility attached to an energy asset.
In other words, the same AI wave that strips the public miners of their Bitcoin-first identity is also dismantling the financialized consolidation of the mining industry. The bull market centralized mining because only public balance sheets could buy power at scale. The AI wave is pushing those same balance sheets into another industry, leaving room for smaller actors who mine for reasons other than satisfying the quarterly expectations of their shareholders. We may look back on the AI-forced exodus as the moment Bitcoin mining became more decentralized, not less. The giants who absorbed the risk are becoming real estate trusts for someone else's compute. The ethos of mining belongs again to the edge of the grid, where no quarterly report can reach.
I do not hold this view naively. There is a darker version of the future in which the residual niche โ the curtailment niche, the off-peak niche, the stranded-energy niche โ turns out to be too small to sustain the network's long-term growth, and Bitcoin's security budget plateaus at a level that remains safe but loses the margin of excess it once enjoyed. There is also a version where a future generation of AI infrastructure becomes as flexibly interruptible as mining, erasing Bitcoin's comparative advantage as the grid's preferred sponge. The market separation is not a final resting state. It is an equilibrium that lasts only as long as the underlying cost structures justify it. Anyone who claims certainty in either direction is selling something.
But the contrarian position must be stated honestly. The question the article posed โ will AI push Bitcoin mining out of the market โ presupposes that the market is one unified arena where two combatants fight over the same prize. In truth, we are watching the opposite. The market is sorting itself into two arenas divided by the physics of load flexibility. The prize was never the same to begin with. AI wants the grid's promise. Bitcoin mining wants the grid's silence.
What to Watch While the Separation Completes
For those of us who survived long enough to read the raw numbers, the months ahead offer four signals worth more attention than the price of Bitcoin itself. First, the hashrate floor. If the network holds its current territory and reclaims the path toward a zettahash, the difficulty adjustments are doing their work faster than capital exits. If it sinks below 800 exahashes, the bear case โ not for Bitcoin's survival but for the economics of marginal miners โ gains new credibility. Second, the public miners' treasury. If the Q1 record sale of 32,000 coins proves to be a one-time recapitalization event and holdings stabilize, the selling pressure that helped pin prices near $74,000 to $80,000 will fade like a tide that finally stops going out. Third, the execution quality of the AI leases. TeraWulf and Hut 8 are setting the standard; Core Scientific's lawsuits remind us that a press release is not a delivery date. Fourth, the energy producers. When more companies like ENGIE announce mining pilots, we will know the flexible-load thesis has moved from white paper to construction schedule.
The soul does not mint; it manifests. What Bitcoin mining is manifesting right now is not its destruction but its usefulness at the most unexpected boundary: the boundary where energy is abundant and attention is cheap enough to point a machine at a problem. If the next cycle of mining is powered not by speculative capital but by otherwise wasted sunlight in the Brazilian interior, then the industry did not lose its war with AI. It simply realized it was fighting on the wrong front.
The question was never whether AI would push Bitcoin mining out of the market. It was whether mining was too dependent on a market at all. The machines that remain after this exodus will not be the ones that dreamed the loudest. They will be the ones that found a home where the grid forgot its own excess. Listen, and you can hear the difficulty adjust โ a slow, certain heartbeat beneath the noise โ still keeping time after all these years.