Bitcoin's network hashrate has spent 320 consecutive days below its October 2025 record of 1,157 EH/s, the longest such stretch in a decade. The seven-day average sat near 914 EH/s at the end of August 2026, a roughly 21% decline from peak. Mining difficulty has fallen 19.3% from its all-time hig...
"Bitcoin is no longer a long-term strategic focus." — Asher Genoot, CEO of Hut 8, Q2 2026 Earnings Call
Bitcoin's network hashrate has spent 320 consecutive days below its October 2025 record of 1,157 EH/s, the longest such stretch in a decade. The seven-day average sat near 914 EH/s at the end of August 2026, a roughly 21% decline from peak. Mining difficulty has fallen 19.3% from its all-time high of 155.97 trillion to 125.81 trillion as of the September 6 adjustment — a year-over-year decline that has occurred only twice in Bitcoin's history.
The cause is structural, not cyclical. Publicly traded miners have signed more than $30 billion in long-term artificial intelligence data center leases since mid-2025. With Bitcoin at approximately $77,100 and CoinShares estimating weighted average cash production costs near $80,000 per coin, miners face negative operating margins on every block mined. The opportunity cost calculus has shifted: a megawatt of power locked into a 15-year AI lease with an investment-grade counterparty now yields more predictable revenue than a megawatt pointed at SHA-256 computation against a 3.125 BTC block subsidy.
The result is a bifurcated industry. Miners that pivoted early — Core Scientific, TeraWulf, Hut 8 — trade at premiums to book value. Those that stayed the course — MARA, CleanSpark — report double-digit revenue declines and widening losses.
Bitcoin's network first crossed 1 zettahash per second (1,000 EH/s) in late August 2025 and peaked at approximately 1,157 EH/s in October 2025. It has not returned. As of September 7, 2026, the network operates at roughly 900 EH/s, according to data aggregator CoinWarz.
The difficulty metric tells a parallel story. Bitcoin logged ten difficulty drops versus seven increases across 2026's seventeen adjustment epochs through September 6. The latest adjustment on September 6 at block 965,664 brought difficulty to 127.45 trillion, a modest 1.31% increase — but still 19.3% below the October 2025 all-time high of 155.97 trillion.
According to Blockspace Media, this marks only the second time in Bitcoin's history that mining difficulty has fallen on a year-over-year basis. The first occurred during the 2022 bear market. Approximately 150 EH/s remains sidelined — equivalent to roughly 15% of the network's peak capacity.
An acute demonstration of the network's vulnerability to external shocks came in late January 2026, when Winter Storm Fern knocked hashrate from 1.1 ZH/s to 663 EH/s — a 30-40% plunge in days — as power outages swept the southern United States.
The math facing Bitcoin miners in September 2026 is straightforward and unfavorable.
Hashprice — daily revenue per petahash per second of deployed capacity — hovers near $29/PH/s/day as of July 2026, according to Hashrate Index. This is near or below breakeven for the majority of deployed hardware. The figure reflects the combined pressure of the April 2024 halving (which cut the block subsidy from 6.25 to 3.125 BTC) and sustained difficulty levels that, while down from peak, remain historically elevated.
Production costs have risen in lockstep. CoinShares' Q1 2026 mining report pegged the weighted average cash cost at $79,995 per bitcoin in Q4 2025, with hash price falling further to $29/PH/s/day in Q1 2026. When non-cash expenses including depreciation and stock-based compensation are factored in, all-in costs reached approximately $137,000 per BTC as of early 2026 — well above the current spot price near $77,100.
CoinShares warned in its Q1 2026 report that 15-20% of publicly listed Bitcoin miners are now operating unprofitably on a cash-cost basis alone.
Publicly traded miners sold more than 32,000 BTC in Q1 2026 alone, exceeding their combined sales for all of 2025, according to data tracked by insights4.vc. The selling was not panic liquidation — it was operational necessity, as miners needed fiat to fund the capital-intensive pivot to AI infrastructure.
The aggregate scale of committed AI infrastructure contracts across the public mining sector now exceeds $30 billion. The deals are characterized by long duration (10-20 year leases), investment-grade counterparties, and fixed-rate revenue — a profile that contrasts sharply with the volatile, subsidy-dependent economics of Bitcoin mining.
Key contracts announced through September 2026:
| Company | AI Contract Value | Counterparty | Duration | |---------|------------------|--------------|----------| | TeraWulf | $12.8B | Fluidstack (Google-backed), others | Multi-year | | Hut 8 | $9.8B (Texas facility) | Undisclosed | 15 years | | Core Scientific | $10B+ | CoreWeave | Multi-year | | CleanSpark | $6.6B | Undisclosed | 20 years |
TeraWulf has been the most explicit about its trajectory. Management stated its intention to exit Bitcoin mining operations entirely by end of 2026, making it the first major public miner to fully abandon SHA-256. In Q2 2026, HPC lease revenue from Anthropic and Fluidstack deals reached $31.9 million — 71% of total revenue — while crypto mining revenue fell to $12.8 million.
Hut 8 CEO Asher Genoot repositioned the company around "integrated power and compute rather than merchant mining." The company's total contracted AI portfolio reached $26.6 billion across all facilities and pipeline, according to its 2026 investor presentations.
The mining sector has split into two cohorts, and the market is pricing the divergence accordingly.
AI-First Miners (outperforming):
Mining-Dependent Operators (underperforming):
The valuation gap between these cohorts reflects a market judgment about the durability of each revenue stream. Fixed-rate AI leases with creditworthy counterparties command premium multiples; volatile mining revenue tied to halving cycles and electricity arbitrage does not.
A 21% hashrate decline from peak raises a question that the Bitcoin community has historically dismissed but can no longer avoid: at what point does reduced computational security become a material concern?
The economic cost of a sustained 51% attack scales linearly with hashrate. At 900 EH/s, the theoretical cost of renting sufficient hardware for a 51% attack is lower than at 1,157 EH/s by roughly the same 21% margin. According to a KuCoin research note, this decline "directly reduces the economic cost of a 51% attack, thereby compromising the base layer's security."
In practice, several mitigating factors apply. The hashrate remains globally distributed — the US holds 37.4% of total capacity, Russia 16.9% — making coordinated hardware commandeering logistically impractical. No chain disruptions have been observed. The absolute cost of sustaining a 51% attack for any meaningful duration still exceeds plausible economic incentives.
However, the directional trend matters. If miners continue to reallocate megawatts from SHA-256 to AI workloads, the network's security budget — funded by the block subsidy and transaction fees — must compensate through higher BTC prices, increased transaction fee revenue, or both. Neither is guaranteed.
The composition of Bitcoin mining's energy consumption is evolving alongside the capacity decline.
According to CBECI data, 52.4% of Bitcoin mining electricity now comes from zero-emission sources, up from 37.6% in 2022. The breakdown: hydropower at 23.4%, wind at 15.4%, nuclear at 9.8%, and solar at 3.2%. Bitcoin's annualized consumption stands at approximately 170-180 TWh as of February 2026, representing 0.7-0.8% of global electricity production.
Nuclear's share has grown from 4% in 2021 to nearly 10% in 2025. TeraWulf's early joint venture with Talen Energy near Pennsylvania's 2.5 GW Susquehanna nuclear plant — drawing power at $0.02/kWh — provided a template that other operators have replicated.
The irony is structural: the miners with the cheapest, cleanest power are the same ones best positioned to pivot to AI, because hyperscalers demand both low-cost electricity and sustainability commitments. The power assets that made mining profitable in 2023 now make AI hosting more attractive in 2026.
The Bitcoin mining industry's transformation from a monolithic hash-producing sector into a specialized tier of the AI infrastructure market is no longer speculative — it is being measured in quarterly earnings, signed lease agreements, and declining hashrate charts.
The economic logic is clear. A megawatt of power committed to a 15-year, fixed-rate AI lease with Google-backed Fluidstack or CoreWeave generates more predictable returns than the same megawatt directed at mining Bitcoin against a halved block subsidy and $80,000+ production costs. The market agrees: mining-to-AI converts trade at premiums, while pure miners face analyst downgrades.
What remains unresolved is the long-term equilibrium. Bitcoin's security model depends on miners finding it profitable to deploy hashrate. If the AI opportunity continues to siphon capacity, the network's security budget must adjust — through higher Bitcoin prices, increased transaction fees, or protocol-level changes. None of these adjustments is automatic.
For now, the data shows an industry in transition, not in crisis. Hashrate has declined but remains at levels that dwarf anything seen before 2024. The network functions normally. But the structural incentive shift is real, and its implications for Bitcoin's long-term security model warrant monitoring beyond the current earnings cycle.