Over the past 316 days, Bitcoin's hashrate has failed to set a new all-time high. The network peaked at 1.1 zettahash per second in October 2025. Since then, it has retreated multiple times to 900 exahash. Mining difficulty dropped 19.1% from its record of 155.97 trillion to 126.23 trillion by June 2026. Hashprice — the daily revenue per petahash — now sits at $31.8, down 50% from the $63 peak in July 2025. The narrative is clear: AI is squeezing miners out of the power market. But the data tells a more granular story — one of structural segmentation, not extinction.
Context: The Protocol Hasn't Changed
Bitcoin's proof-of-work consensus remains untouched. The difficulty adjustment algorithm still recalibrates every 2,016 blocks. Miners using ASICs cannot switch to AI workloads because the hardware is incompatible — SHA-256 chips cannot run neural networks. The shift is not about technology substitution but about resource allocation. Miners own power contracts, substations, and cooling infrastructure. AI operators need those same assets but with higher reliability and fiber connectivity. The tension is not between two technologies but between two business models: volatile mining revenue versus stable data center lease income.
Trust no one, verify the proof, sign the block. This principle applies not just to transactions but to market narratives. The claim that "AI will push mining out of the market" collapses when we examine the numbers.
Core: The Economics of Segmentation
Let's dissect the miner P&L. At $31.8/PH/s/day, only the most efficient hardware survives. A Bitmain Antminer S21 XP (13.5 J/TH) breaks even at $0.088–$0.124/kWh. Older S19 units (25+ J/TH) need power under $0.055/kWh — a price only available in stranded energy markets. The result is a natural split:
| Metric | Value | Implication | |--------|-------|-------------| | Hashprice | ~$31.8/PH/s/day | Down 50% from 2025 peak | | Break-even for S21 XP | $0.124/kWh at $77k BTC | Profitable at efficient sites | | Break-even for S19 | $0.055/kWh | Only viable with near-free energy | | Difficulty drop | 19.1% from peak | Third largest in ASIC era | | Public miner BTC sales | 32,000 BTC in Q1 2026 | Record quarterly sell-off |
Core Scientific reported a -56% gross margin on self-mining in early 2026. Yet its data center leasing business generated nearly $80 million in gross profit. TeraWulf now derives 71% of revenue from HPC/AI hosting. These are not mining companies pivoting to AI — they are energy infrastructure companies that happened to start with Bitcoin. The mining revenue becomes the floor, not the ceiling.
Trust no one, verify the proof, sign the block. The proof here is the divergence between miner stock performance and BTC price. Core Scientific stock rose 75.8% in H1 2026 while Bitcoin traded sideways. The market is pricing miners as data center REITs, not Bitcoin proxies.
Contrarian: The Real Danger Is Execution, Not Competition
The prevailing fear is that AI will outbid miners for power, driving them out of business. I see a different risk: miners who over-leverage into AI transition and fail to deliver. Core Scientific secured $3.3 billion in junk bonds to fund its pivot. It now faces a class-action lawsuit alleging it overstated AI demand and concealed construction delays. TeraWulf, by contrast, backed its AI contracts with investment-grade counterparties like Google-cloud-backed Fluidstack. The difference is execution quality, not market size.
Meanwhile, a quieter trend is emerging: energy producers entering mining directly. ENGIE, the French utility giant, is evaluating Bitcoin mining on its Brazilian solar farms to absorb curtailed power. This flips the script. Instead of miners competing for grid electricity, mining becomes a flexible load for renewable assets. The hardware can shut down in seconds when the grid needs power — AI datacenters cannot. This gives mining a unique value proposition that AI cannot replicate.
The contrarian truth: AI will not destroy mining. It will force a separation. Premium power (stable, high-capacity, fiber-rich) goes to AI. Stranded power (intermittent, remote, low-cost) stays with mining. The miners that survive are those that either execute flawlessly on AI hosting or embrace their role as energy recyclers.
Trust no one, verify the proof, sign the block. The proof here is the difficulty adjustment mechanism itself. Every time miners shut down, difficulty drops, profitability recovers for the survivors. This is not a death spiral — it is a self-correcting market.
Takeaway: Watch the Hash, Not the Hype
The question is not whether AI pushes mining out of the market. It is whether the market will split into two tiers: high-reliability AI power and low-cost mining power. If BTC price stays below $74,000, more efficient miners will exit, hashrate may drop to 800 EH/s, and difficulty will adjust accordingly. If AI transition fails for leveraged miners, we will see forced liquidations of both hardware and BTC reserves.
But the network will survive. Bitcoin's security depends on economic incentives, not corporate strategies. The hash will follow the cheapest electrons. And those electrons are increasingly found in places AI cannot reach.
The next 12 months will separate the operators who understand their core asset — power contracts, not hashrate — from those who chase narratives without execution. Code does not forgive. Neither does the market.
