Bitcoin's network hashrate reclaimed 1 zettahash per second on October 16, 2026 — its highest reading since March — even as publicly traded miners accelerated their exit from the sector. Mining difficulty jumped 5.92% to 140.58 trillion, the third-largest increase of the year. The next adjustment...
"Bitcoin mining is a high-risk, variable-reward business compared to AI data center hosting, which typically offers more stable, long-term contracts." — CoinShares, Q2 2026 Bitcoin Mining Report
Bitcoin's network hashrate reclaimed 1 zettahash per second on October 16, 2026 — its highest reading since March — even as publicly traded miners accelerated their exit from the sector. Mining difficulty jumped 5.92% to 140.58 trillion, the third-largest increase of the year. The next adjustment, expected October 28, could add another 14.24% if block production maintains its current pace.
The numbers reveal a paradox. Hashrate is rising while the companies that built it are leaving. Core Scientific now derives 83% of revenue from AI colocation. TeraWulf reports 71% from HPC leases. IREN is decommissioning mining hardware. Yet the network has never been more computationally dense. The explanation lies in a migration of hashrate from public markets to private operators, many running sub-$0.04/kWh power in regions where disclosure requirements do not apply.
This report examines the economics of Bitcoin mining 30 months after the April 2024 halving: who can still profit, who cannot, and what the structural shift toward AI infrastructure means for the network's long-term security budget.
Bitcoin's seven-day average hashrate crossed 1,000 EH/s (1 ZH/s) on October 16, 2026, a level last seen on March 3. The recovery occurred despite approximately 75 EH/s leaving the network in the first half of the year, according to CoinShares, as operators redirected power capacity toward AI and high-performance computing workloads.
Mining difficulty adjusted to 140.58 trillion at block height 971,712 on October 16 — a 5.92% increase and the third-largest single adjustment of 2026, according to data from multiple block explorers. Early estimates for the October 28 adjustment point to a potential 14.24% increase if current block production rates hold.
The recovery signals a shift in who is mining. Public miners, which once accounted for roughly 28-30% of global hashrate, are systematically reducing their mining footprint. Private operators — many based in jurisdictions with lower electricity costs and fewer reporting obligations — are filling the gap. The hashrate is growing; the shareholder-visible portion of it is shrinking.
The weighted average ex-tax cash cost to produce one bitcoin among listed miners stood at approximately $75,500 in Q2 2026, according to CoinShares' quarterly mining report. JPMorgan's independent estimate placed the figure at $78,000. Both numbers sit uncomfortably close to Bitcoin's Q2 closing price of $58,400 — meaning the listed mining sector operated below cash breakeven in aggregate during the quarter.
Cost dispersion across the industry is extreme. American Bitcoin Corp (ABTC) recorded the lowest cash cost at $43,851/BTC. CleanSpark produced at $71,995/BTC. Riot Platforms came in at $74,955/BTC. At the other end, TeraWulf's self-mining allocation reached $134,446/BTC, reflecting the company's deliberate wind-down of mining operations in favor of HPC leasing.
Electricity remains the dominant variable. An efficient operation running Antminer S21 Pro hardware at $0.045/kWh faces a direct electricity cost of $35,000-$40,000 per BTC. Add hardware depreciation, facilities, cooling, and overhead, and the all-in figure roughly doubles. Operations paying above $0.07/kWh are losing money on most hardware generations at current prices.
With Bitcoin trading near $80,000-$83,000 in October, the network has crossed back above the average production cost. Margins exist again — but they are thin, and they are not evenly distributed.
September 2026 delivered $1.12 billion in total miner revenue, an 11% increase over August and the strongest monthly payout since January, according to on-chain data. Daily miner revenue climbed from approximately $27 million at July's lows to as much as $48 million — a 78% increase, per CryptoQuant.
Hashprice — the dollar revenue per petahash per day — recovered to above $40 in September, its highest level since May. As of October 18, hashprice had settled to approximately $35.34/PH/day, reflecting the upward difficulty adjustment that redistributed the same block rewards across more competing hashrate.
The recovery was driven almost entirely by BTC price appreciation. Bitcoin's move from the $58,000-$60,000 range in June-July to $80,000+ in October mathematically increased the dollar value of each 3.125 BTC block reward. Transaction fees contributed almost nothing: over the 30 days ending October 6, miners collected roughly 87.6 BTC in fees out of 13,588 BTC in total rewards — approximately 0.64%.
For operators running sub-15 J/TH hardware at $0.07/kWh hosting costs and a BTC price near $83,000, margins range from 42% to 60%. Earlier in the year, when hashprice hit an all-time low of $27.7/PH/day in June, many of these same operations were cash-flow negative.
The most significant structural change in the mining sector is not a mining development at all. It is the systematic conversion of Bitcoin mining infrastructure into AI data centers.
The numbers are large. Applied Digital signed approximately $11 billion in contracts with CoreWeave. Core Scientific committed $10.2 billion with the same counterparty. IREN holds a $9.7 billion multi-year partnership with Microsoft. TeraWulf entered a $9.5 billion arrangement with Fluidstack. These are long-term, contracted revenue streams — a different economic model from the daily-settlement, spot-price-dependent reality of Bitcoin mining.
The revenue shift is already visible in quarterly filings. Core Scientific generated $136.7 million in AI colocation revenue in Q2 2026, compared to $27.5 million from Bitcoin mining — an 83:17 ratio. TeraWulf reported $31.9 million from HPC leases versus $12.8 million from mining. IREN is on track to derive approximately 70% of revenue from AI by year-end.
The financial logic is straightforward. CoinShares estimates that AI infrastructure generates multiples of traditional mining profits per megawatt. IREN's Microsoft partnership alone projects approximately $1.94 billion in annualized revenue at an estimated 85% project-level EBITDA margin. No Bitcoin mining operation approaches those unit economics.
Meanwhile, the listed miners' balance sheets show the toll. MARA Holdings and CleanSpark posted $851.1 million in combined net losses for Q2 2026, with Bitcoin fair-value markdowns accounting for roughly $459 million of that. MARA's revenue fell 27% year-over-year to $174.9 million. CleanSpark's dropped 30.5% to $138 million.
The stock market has noticed the divergence. Some mining stocks have outperformed Bitcoin itself — not because mining is profitable, but because investors are repricing these companies as AI infrastructure plays. The hashrate these companies operated still runs; it has simply changed owners.
Transaction fees now represent just 0.69% of total miner revenue — approximately $1,000 per block against a $200,000 block reward, according to TradingView data through October 6. This is the lowest ratio in a decade.
The implication is structural. Bitcoin's security model depends on miners being compensated sufficiently to maintain honest hashrate. The block subsidy — currently 3.125 BTC — halves again in 2028 to 1.5625 BTC. Unless transaction fees increase substantially, each halving directly reduces the economic incentive to mine honestly.
At current fee levels, eliminating the block subsidy entirely would reduce miner revenue by 99.3%. The network currently pays approximately $14.4 million per day in block subsidies and roughly $100,000 per day in fees. This ratio must change — either through higher on-chain transaction volume, higher fee rates per transaction, or the development of new fee-generating use cases — for the post-subsidy security model to function.
Bitcoin's annualized electricity consumption stands at an estimated 170-180 TWh, according to the Cambridge Bitcoin Electricity Consumption Index, representing approximately 0.7-0.8% of global electricity production. That consumption must be economically justified by ongoing revenue. At 0.69% fee share, the network's long-term energy bill is subsidized almost entirely by inflation — a model with a mathematical expiration date.
The current generation of mining hardware centers on Bitmain's Antminer S21 series, which spans eight models with hashrates between 190 and 473 TH/s and energy efficiency between 19 and 11 J/TH.
The top-performing unit — the S21 XP Hyd (hydro-cooled) — delivers 473 TH/s at 12.0 J/TH. The air-cooled S21 XP achieves 270 TH/s at 13.5 J/TH. The budget-oriented T21 runs at 190 TH/s and 19 J/TH — viable only at sub-$0.05/kWh power costs.
Auradine has announced next-generation ASICs targeting sub-10 J/TH efficiency by Q3 2026, which would represent a generational improvement. Each step down in joules-per-terahash directly reduces the electricity cost per bitcoin produced. A 10 J/TH machine at $0.05/kWh would produce Bitcoin at roughly 20-25% lower electricity cost than today's 13.5 J/TH air-cooled leaders.
The hardware cycle creates a recurring pattern: new machines improve margins temporarily, then difficulty adjusts upward as more efficient hashrate enters the network, compressing margins back toward breakeven. The equilibrium point is determined by the marginal producer — the least efficient machine that can still cover its electricity costs at prevailing BTC prices.
Bitcoin mining in October 2026 presents a sector defined by bifurcation. The network itself is computationally stronger than ever. The companies that built it are leaving. The hashrate persists, migrating from quarterly-reporting public companies to private operators willing to accept thinner margins and longer payback periods.
The AI pivot is not a sideshow — it is a structural repricing of what power infrastructure is worth. A megawatt dedicated to Bitcoin mining generates variable, price-dependent revenue. The same megawatt leased for AI workloads produces contracted, multi-year cash flows at higher margins. The economic choice, for any operator with the technical capability to make the switch, is unambiguous.
What remains is the security question. If the most sophisticated, best-capitalized operators leave mining for AI, and the fee share of miner revenue remains below 1%, the network's long-term security model rests on an assumption that BTC price appreciation will indefinitely compensate for subsidy halvings. That is a bet, not a guarantee. The data supports monitoring this closely.