Bitcoin's mining difficulty has spent 322 consecutive days below its October 2025 record of approximately 156 trillion, the longest sustained drawdown since 2012. As of September 6, 2026, difficulty stands at 127.45 trillion — 18.3% below peak — after the network's 15th adjustment of the year. Th...
"We are no longer a Bitcoin company." — Ben Gagnon, CEO, Bitfarms (now Keel Infrastructure)
Bitcoin's mining difficulty has spent 322 consecutive days below its October 2025 record of approximately 156 trillion, the longest sustained drawdown since 2012. As of September 6, 2026, difficulty stands at 127.45 trillion — 18.3% below peak — after the network's 15th adjustment of the year. The decline was not triggered by a government ban or a natural disaster. It is a capital allocation decision: operators are voluntarily redirecting megawatts away from SHA-256 hashing and toward artificial-intelligence infrastructure.
Public miners liquidated a record 32,000 BTC in Q1 2026, exceeding the combined net sales of all four quarters in 2025 and surpassing the roughly 20,000 BTC dumped during the Terra-Luna collapse in Q2 2022. The proceeds are funding a structural pivot. According to CoinShares' Q1 2026 mining report, publicly listed miners could derive up to 70% of their revenues from AI by December 2026, up from approximately 30% at the start of the year. Miners with secured HPC contracts now trade at EV/NTM sales multiples of 12.3x; pure-play miners trade at 5.9x.
The result is a two-speed mining industry. Companies that locked in power purchase agreements and data-center shells before the AI demand wave are converting those assets into high-margin GPU hosting contracts. Companies that did not are selling bitcoin reserves to cover electricity bills.
Bitcoin's difficulty algorithm adjusts every 2,016 blocks — roughly every two weeks — to maintain a target block time of approximately ten minutes. When hashrate falls, difficulty falls with it. The current drawdown from the October 2025 peak of ~156 trillion ranks as the third deepest since application-specific integrated circuits (ASICs) became the standard mining hardware, behind only the aftermath of China's 2021 mining ban and the 2018 bear market contraction.
In 2026 alone, the difficulty has recorded ten decreases totaling -45.27% and eight increases totaling +33.34%, producing a net decline of approximately 11.93% year-to-date. The September 6 adjustment at block height 965,664 brought difficulty to 127.45 trillion, a modest +1.31% increase, but still well below peak.
The current drawdown has lasted 322 days from peak as of mid-September 2026. The prior comparable event — the 2021 China ban — produced a sharper but shorter decline: difficulty collapsed 28% in weeks before recovering within four months as displaced miners redeployed in Kazakhstan, the United States, and other jurisdictions. The 2026 drawdown is different in character. It is gradual, voluntary, and driven by economics rather than regulation.
Bitcoin's network hashrate briefly touched 1,001 EH/s on September 4, 2026, per on-chain data trackers, flirting with the 1 zettahash milestone. However, the 7-day moving average stood at approximately 915 EH/s at the end of August, and the 30-day average near 912 EH/s. As of September 7, the network was running at roughly 900 EH/s.
These figures represent a meaningful decline from the late-2025 peak near 1 ZH/s. The gap between daily spikes and sustained averages reflects the intermittent nature of the remaining hashrate: some large operators cycle machines on and off in response to real-time electricity pricing and hashprice signals.
Hashprice — the dollar revenue per petahash per second per day — rose 22.24% over the month preceding the September adjustment, climbing from $32.42 to $39.63, according to Bitcoin.com. This improvement was driven primarily by bitcoin's price recovery rather than a reduction in competition. Bitcoin's price dictates revenue; difficulty dictates cost. When price rises faster than difficulty, margins temporarily expand, drawing marginal hashrate back online and setting up the next upward adjustment.
Public Bitcoin miners sold over 32,000 BTC in Q1 2026, according to on-chain data compiled by multiple research firms. This figure surpassed full-year 2025 net sales and exceeded the roughly 20,000 BTC liquidated during the Terra-Luna crisis in Q2 2022.
Major sellers included:
| Company | BTC Sold (Q1 2026) | Estimated Proceeds | |---------|--------------------|--------------------| | MARA Holdings | 15,133 BTC | ~$1.1 billion | | Riot Platforms | 3,778 BTC | ~$289.5 million | | Core Scientific | ~1,900 BTC | ~$175 million (Jan. only) |
According to CoinDesk, Core Scientific sold $208 million of bitcoin in Q1 2026 as its AI pivot continued. MARA, Genius Group, and Nakamoto Holdings together disclosed sales of more than 15,000 coins in a single week during Q1. These were not routine sales of freshly mined production to cover electricity bills. They were strategic drawdowns of treasury reserves.
The capital raised is flowing into GPU infrastructure, data-center construction, and AI hosting contracts. This represents a one-directional flow: bitcoin reserves are being converted into physical infrastructure assets that cannot be easily reversed.
Bitcoin's April 2024 halving reduced the block reward from 6.25 to 3.125 BTC. For public miners, the weighted average cash cost to produce one bitcoin rose to approximately $79,995 by Q4 2025, according to CoinShares. This figure includes direct electricity costs, hosting fees, and SGA but excludes depreciation and financing costs.
All-in production costs range from $38,000 to $92,000 per bitcoin in 2026, depending on electricity rates and hardware efficiency. The threshold for profitability sits at approximately $0.08/kWh for operators running current-generation ASICs. Electricity accounts for 60-80% of total mining costs.
Gross margins in bitcoin mining have compressed from approximately 90% during the 2021 cycle peak to around 60% in 2026. By contrast, AI cloud infrastructure generates margins of approximately 85% with lower energy overhead per dollar of revenue. The economic incentive to redirect megawatts is quantifiable: the same power infrastructure generates higher margins running GPUs than running SHA-256 ASICs at current bitcoin prices and difficulty levels.
The migration from SHA-256 hashing to GPU hosting is no longer a marginal diversification play. It is becoming the primary business for several of the largest public miners.
Core Scientific (CORZ): Colocation revenue for AI data centers reached $77.5 million in Q1 2026, surpassing crypto mining revenue of $30.1 million. CoreWeave accounted for approximately 77% of total revenue in H1 2026. The company holds a 590 MW, $10.2 billion, 12-year contract with CoreWeave and paid $42 million to exit a Bitcoin mining deal with Block in order to free up capacity for AI workloads. Management confirmed bitcoin mining will continue winding down through 2026, with only one or two sites expected to remain operational by year-end.
Riot Platforms (RIOT): Recorded $33.2 million in AI hosting revenue in its first quarter of data-center revenue, which CEO Jason Les described as an "inflection point." Riot possesses 1.7 GW of approved power capacity. Its AMD lease is expected to generate $311 million in initial revenue and up to $1 billion with extensions.
Bitfarms → Keel Infrastructure: Announced in February 2026 that it would redomicile from Canada to the United States and rebrand as Keel Infrastructure, explicitly exiting the bitcoin mining identity. Shareholders approved the move in March 2026.
Aggregate contracts: According to CoinShares, Bitcoin miners have signed GPU colocation and cloud service deals with hyperscalers worth over $70 billion in aggregate over the course of 2025 and early 2026.
The market is pricing the AI pivot selectively. Bernstein raised price targets on Riot Platforms to $30 from $25 and Core Scientific to $32 from $24 — both rated Outperform — while cutting MARA Holdings to $17 from $23 with a Market Perform rating.
The divergence reflects a clear analytical framework: miners that have signed binding HPC contracts with creditworthy counterparties are valued as infrastructure companies. Miners that remain dependent on bitcoin block rewards and transaction fees are valued on hashrate economics — which are currently unfavorable.
Stock performance in 2026 reflects this split. During the April-to-June rally, RIOT gained 83%, HUT gained 72%, BITF gained 50%, and CORZ gained 31%. The difference in performance tracks directly to each company's AI revenue mix and contracted backlog.
MARA, which has not pivoted as aggressively, reported Q1 2026 revenue down 18% year-over-year to $174.6 million with a net loss of $1.3 billion and diluted loss per share of $3.31.
The sustained difficulty drawdown raises a structural question about Bitcoin's security model. The network's security budget — the sum of block rewards and transaction fees paid to miners — is effectively shrinking in real terms as the halving cycle continues. If miners continue to redirect capacity to AI, the network's hashrate becomes more dependent on bitcoin's price sustaining levels that justify the opportunity cost of running ASICs instead of GPUs.
The counterargument is that difficulty is self-correcting by design: as hashrate falls, difficulty falls, making mining cheaper for remaining operators, which stabilizes the network. The September data supports this — hashprice rose 22% as difficulty eased, pulling marginal hashrate back online. But the structural trend is clear: mining bitcoin is increasingly competing against AI infrastructure for the same scarce resource — permitted, grid-connected power capacity in favorable jurisdictions.
The Bitcoin mining industry is undergoing a capital reallocation event that has no precedent in its 17-year history. Previous difficulty drawdowns were caused by external shocks — regulatory bans, exchange collapses, or bear markets. The 2026 drawdown is voluntary. Miners are not being forced off the network by losses alone; they are choosing to deploy the same infrastructure — power, cooling, network connectivity, physical shells — toward higher-margin workloads.
The difficulty algorithm will continue to adjust downward as hashrate exits, making mining incrementally more profitable for those who remain. This is the mechanism working as designed. But the broader implication is that Bitcoin's security model is now in direct competition with AI compute demand for the same physical infrastructure. The companies that once anchored the network's hashrate are becoming AI hosting companies that happen to mine some bitcoin on the side.
Whether this matters for Bitcoin's long-term security depends on whether price appreciation can sustainably offset the opportunity cost. The data, as of September 2026, shows the market has not yet resolved that question.