Hashprice hit $23.9 per PH/s per day in early 2026. That is the lowest level since the 2018 bear market. At the same time, Core Scientific stock surged 75% in the first half of 2026. The disconnect is not a market inefficiency. It is a signal. The market is pricing in a structural shift: mining companies are no longer pure plays on Bitcoin’s price. They are becoming energy infrastructure REITs with an AI twist. But the real question is not whether AI will push mining out of the market. It is whether the mining industry will split into two irreconcilable camps—one that dies, one that thrives.
The numbers are brutal for the status quo. Bitcoin’s network hashrate peaked above 1.1 ZH/s in October 2025. By August 2026, it had retreated to around 900 EH/s—a drop of roughly 18%. Mining difficulty followed: from a high of 155.97T down to 126.23T in the same period, a 19% decline. That is the third-largest difficulty drop in the ASIC era. The mechanism worked exactly as coded. Miners turned off unprofitable machines; the protocol automatically reduced difficulty to restore equilibrium for the survivors. But equilibrium at 900 EH/s is not the same as equilibrium at 1.1 ZH/s. The composition of that hashrate has changed, and will continue to change.
The core technical constraint is simple and often ignored: ASIC and GPU are not interchangeable. An Antminer S21 can only compute SHA-256 hashes. It cannot train a large language model. A miner cannot flip a switch and start serving AI workloads on the same hardware. When a mining company pivots to AI hosting, it means building or retrofitting entirely new data centers with different power, cooling, and networking requirements. The old mining hardware becomes stranded—or is sold to secondary markets in jurisdictions with lower electricity costs. This is not a technology upgrade. It is a capital reallocation.
I have seen this pattern before. In 2017, during the ICO frenzy, I led a six-person audit of the 2x Capital smart contracts. We found an integer overflow in their leverage calculation logic. The code allowed a user to borrow more than the pool could cover under extreme volatility. The fix was a simple check. But the underlying problem was not the code—it was the assumption that leverage could grow forever without a safety valve. Bitcoin mining has the same problem. The leverage is not financial leverage. It is operational leverage: the assumption that hashrate can grow forever because the block subsidy will always be worth more. That assumption broke when AI data centers started bidding up power prices.
Hashprice compression is not a temporary cycle. It is a structural squeeze. The hashprice fell from about $63/PH/s/day in mid-2025 to $31.8 in August 2026. At those levels, only the most efficient ASICs can break even at typical industrial electricity rates of $0.088–$0.124/kWh. Older generation machines, like the S19 series (25+ J/TH), need power below $0.055/kWh to survive. Those machines are die—ing. The result is a forced upgrade cycle that favors large, well-capitalized miners with access to cheap power. The small players with older hardware are being squeezed out.
But the story does not end with mining becoming unprofitable. It ends with mining bifurcating into two distinct markets. The first market is high-value, high-cost mining in locations with stable, cheap power. This is where AI hosting competes. The second market is opportunistic, low-cost mining on stranded or intermittent energy—solar farms in Brazil that would otherwise curtail production, hydro plants in Paraguay with excess capacity, flare gas sites in Texas. AI cannot use that power because it requires constant, high-quality uptime. Bitcoin miners can. They can turn on and off in seconds to balance the grid. This is the hidden advantage.
The evidence is already there. ENGIE, the French energy giant, is evaluating Bitcoin mining on its Brazilian solar projects. The logic is straightforward: when the sun is strong and the grid cannot absorb all the power, instead of curtailing, the miner runs. When demand spikes, the miner shuts down. The mining hardware becomes a flexible load that turns otherwise wasted energy into revenue. This is not a competitor to AI. It is a complement to renewable infrastructure.
The contrarian angle: AI is not killing mining. It is forcing mining to become more efficient and more decentralized—but in a different direction. The popular narrative says “AI will consume all the cheap power and mining will die.” The data says otherwise. Between Q4 2025 and Q2 2026, publicly listed mining companies saw their share of global hashrate drop by 13.4%. The hashrate that left the public companies did not disappear. It moved to private operators, energy producers, and international players. Bitcoin’s difficulty adjustment ensures that whatever hashrate remains will be profitable. The network does not care who runs the machines. It only cares that the total hashrate is sufficient to secure the chain.
“Code is law, but audit is mercy.” In this context, the code is the difficulty adjustment algorithm. It is immutable. It will always adjust. The mercy is the market: miners who cannot compete on efficiency will exit, but the network will survive. The real risk is not that mining disappears. It is that the hashrate becomes concentrated in a few hands—not via a 51% attack, but via economic consolidation. If only the largest, most efficient miners survive, the network loses some of its geographic and operational diversity. That is a long-term security consideration.
“Infinite yield curves break under finite scrutiny.” The infinite yield curve here is the assumption that mining can grow forever driven by block subsidies and rising Bitcoin prices. Finite scrutiny reveals that block subsidies are fixed (halving every four years) and transaction fees are insufficient to cover operational costs for high-cost miners. The only way to sustain growth is to lower costs or find alternative revenue streams. AI hosting is that alternative stream. But it comes with its own risks.
The execution risk is real. Core Scientific, the poster child of mining-to-AI pivot, is now facing a shareholder lawsuit. The lawsuit alleges that the company overstated its ability to meet AI demand and concealed construction delays. The lawsuit is a reminder that converting a mining facility into an AI data center is not plug-and-play. It requires large capital expenditures, long lead times, and contracts with investment-grade counterparties. TeraWulf, by contrast, has executed well—its HPC leasing business now accounts for 71% of revenue, with Google-backed Fluidstack as a tenant. The market is rewarding execution, not just narrative.
“Composability is leverage until it is liability.” In DeFi, composability means that protocols can stack on top of each other, creating exponential risk. In mining, the composability is between energy contracts, hardware supply chains, and AI service agreements. If one layer fails—say, a power contract is renegotiated upward—the entire business model breaks. The leverage is the debt used to finance the pivot. Core Scientific issued $3.3 billion in high-yield bonds. If AI revenue falls short, that debt becomes a liability that could wipe out equity.
The market is already pricing in the split. Core Scientific’s stock is up 75% year-to-date, while Bitcoin is flat around $77,000. The market is betting that the AI pivot will succeed for some miners, and that those miners will no longer be correlated to Bitcoin’s price. For the miners that fail to execute, the stock will collapse. The mining industry is becoming a game of winners and losers, not a rising tide that lifts all ships.
What does this mean for Bitcoin itself? The network is secure. 900 EH/s is still an astronomical level of security. A 51% attack would cost tens of billions of dollars. The difficulty adjustment ensures that even if hashrate drops another 20%, the remaining miners will be profitable at current prices. The real impact is on the market for mining stocks and on the geographic distribution of hashrate. Expect to see more hashrate move to Latin America, Africa, and the Middle East, where energy is cheap and regulation is friendly. Brazil has already removed import tariffs on efficient mining hardware until 2028. That is a signal.
The takeaway is not about the end of mining. It is about the end of one-size-fits-all mining. The next five years will see two distinct industries emerge: high-cost, high-reliability mining co-located with AI data centers, and low-cost, intermittent mining integrated with renewable energy. The first industry will be dominated by public companies with access to cheap debt. The second will be dominated by energy producers and private operators. The Bitcoin network will benefit from both—more hashrate from stranded energy, more institutional legitimacy from AI-connected miners.
The question every miner must answer is simple: Are you competing on power cost or on execution? If you cannot secure power below $0.05/kWh, you need to pivot to AI hosting. If you can secure power below $0.03/kWh, you should stay pure mining and expand. The middle ground is death. The hashrate floor is not the ceiling. It is the dividing line.
Blind faith is the only true vulnerability. Blind faith that mining would always be profitable. Blind faith that AI would always be the savior. The winners will be those who verify every assumption—power cost, hardware efficiency, counterparty risk—and then build twice. The losers will be those who trust the narrative without auditing the numbers.
