Bitcoin's network hashrate has fallen below 1 zettahash per second for the first time since breaching that threshold in early 2026, driven by a structural migration of mining capacity toward artificial intelligence and high-performance computing (HPC) infrastructure. The protocol has executed six...
"In time, we will have no bitcoin." — Ben Gagnon, CEO, Bitfarms (now Keel Infrastructure)
Bitcoin's network hashrate has fallen below 1 zettahash per second for the first time since breaching that threshold in early 2026, driven by a structural migration of mining capacity toward artificial intelligence and high-performance computing (HPC) infrastructure. The protocol has executed six downward difficulty adjustments this year — a 10.7% cumulative reduction — as publicly listed miners redirect power, capital, and hardware toward AI contracts offering 3–5x the revenue per megawatt-hour of mining.
CoinShares projects that listed miners will derive 70% of revenue from AI/HPC by year-end 2026, up from approximately 30% at the start of the year. More than $70 billion in AI and HPC contracts have been signed across the sector. The weighted average cash production cost per bitcoin among public miners has risen to approximately $90,000, while BTC trades near $76,000 — a negative margin of roughly $14,000 per coin. Miners sold over 32,000 BTC in Q1 2026, a quarterly record, to fund the transition. At least one major miner, Bitfarms, has announced a full exit from bitcoin mining, rebranding as Keel Infrastructure with a 2.2 GW AI data center pipeline.
The implications extend beyond mining economics. Bitcoin's security model, which depends on continuous expenditure of real-world energy, is now in direct competition with AI workloads for the same power capacity. The question is no longer whether the pivot is happening, but whether Bitcoin's fee market can sustain network security as subsidized hashrate withdraws.
Bitcoin's hashrate peaked above 1,000 EH/s (1 ZH/s) in early 2026 before entering a sustained decline. As of mid-May, the network operates between 899 and 970 EH/s, according to CoinWarz data. This represents a year-to-date decline of approximately 4% — the first Q1 drop in six years, following five consecutive years of double-digit growth.
The protocol has responded with six downward difficulty adjustments in 2026. On May 1, difficulty fell 2.3% to 132.47 trillion at block height 947,520. The current difficulty stands at 136.61 trillion following a brief uptick, with the next adjustment estimated on May 29 — projected to decrease again to approximately 134.37 trillion.
Average block time in early May sits at 10 minutes and 28 seconds, slightly above the 10-minute target. The network is running 0.17 minutes slower than expected. While this is well within normal operating parameters, the direction is consistent: hashrate is leaving, and the protocol is mechanically compensating.
Hash price — the revenue earned per petahash per second per day — dropped below $30/PH/s/day in Q1 2026, a five-year low, according to CoinShares. At these levels, 15–20% of legacy mining hardware globally is cash-flow negative.
The math is straightforward. CoinShares' Q1 2026 Bitcoin Mining Report places the weighted average cash production cost per bitcoin among publicly listed miners at approximately $90,000. Bitcoin has traded between $67,000 and $77,000 for much of Q1–Q2 2026. At a production cost-to-price ratio of approximately 1.26, the average listed miner loses roughly 26% on every coin produced.
Efficient operators fare better. MARA Holdings and CleanSpark report all-in production costs in the $34,000–$43,000 range, maintaining 20–50% margins. But these are the exception. The industry-wide all-in cost per BTC sits in a broad range of $75,000–$87,000, according to multiple analyst reports.
The April 2024 halving — which cut the block subsidy from 6.25 BTC to 3.125 BTC — remains the underlying structural pressure. Miners now produce half the bitcoin per block while network difficulty, until recently, continued to rise. Electricity costs have increased year-over-year across most hosting jurisdictions.
The result: a sector that collectively burns cash on its core product.
The alternative is AI infrastructure, and the economics are not close.
AI training clusters pay 3–5x more per megawatt-hour than bitcoin mining generates at current hash prices. AI contracts are multi-year, contracted, and denominated in fiat — a sharp contrast to mining's stochastic block rewards denominated in a volatile asset.
IREN Limited secured a five-year partnership with Microsoft projected to generate $1.94 billion in annualized revenue at an 85% project-level EBITDA margin. Hut 8 signed a $7 billion, 15-year contract with Google-backed Fluidstack for a 245 MW AI data center. These are not speculative figures; they are contracted cash flows backed by hyperscaler counterparties.
Mining companies hold a structural advantage in the competition for AI infrastructure: they already control energized sites with grid interconnection, cooling systems, and power purchase agreements. Traditional data center developers require 3–5 years to bring comparable capacity online. Miners can repurpose existing infrastructure in 12–18 months.
The total addressable opportunity has been sized at $40 billion in annualized revenue by 2026, according to industry estimates. Companies that have completed the pivot report operating margins of 80–90% on AI contracts, compared to breakeven or negative margins on mining operations.
Bitfarms / Keel Infrastructure: The most aggressive exit. CEO Ben Gagnon announced plans to "wind down our Bitcoin mining business in 2026 and 2027." The company sold its 70 MW Paraguay site for up to $30 million, rebalanced to 100% North American operations, and is redomiciling to Delaware under the name Keel Infrastructure. It is developing a 2.2 GW AI/HPC data center pipeline. Bitfarms still holds 1,827 BTC and has stated it will sell "opportunistically into strength."
TeraWulf (WULF): Q1 2026 results show a $427 million net loss, but AI revenue doubled to $21 million — now 60% of total revenue. Bitcoin mining income fell 50% to $13 million. Cash and restricted cash reserves totaled approximately $3.1 billion at quarter-end.
IREN (Iris Energy): Q3 FY2026 (March quarter) revenue was $144.8 million. Bitcoin mining revenue fell to $111.2 million from $167.4 million the prior quarter as hardware was decommissioned ahead of GPU installations. AI cloud services revenue rose to $33.6 million from $17.3 million. Net loss was $247.8 million, impacted by $140.4 million in non-cash impairments on decommissioned mining hardware.
Riot Platforms: Q1 2026 revenue of $167.2 million. Data center business contributed $33.2 million; bitcoin mining revenue fell to $111.9 million from $142.9 million year-over-year. CEO Jason Les described the quarter as an "inflection point" toward becoming a data center operator.
CleanSpark: Beat Microsoft in a bid to operate a 100 MW AI data center in Cheyenne, Wyoming. CEO Matt Schultz stated miners are "uniquely positioned" due to their ability to "build out and energize data centers very rapidly."
The pivot requires capital. Miners are funding AI conversions through a combination of debt issuance, equity raises, and — critically — bitcoin sales.
Publicly traded miners sold over 32,000 BTC in Q1 2026, a record quarterly figure. Key transactions include:
On-chain metrics suggest the peak of miner selling pressure may be passing. The Miner's Position Index (MPI) and Miner Selling Power indicators have recently declined, according to CryptoSlate analysis. However, the structural incentive to sell — funding AI infrastructure with BTC liquidation proceeds — persists.
Bitcoin's security model depends on miners continuously expending energy to produce hashes. The cost to execute a theoretical 51% attack scales directly with hashrate. As hashrate declines, the economic cost of attack decreases proportionally.
The current 4% year-to-date decline is modest in absolute terms. Bitcoin's hashrate remains orders of magnitude above levels that would present a realistic attack surface. The protocol's difficulty adjustment mechanism continues to function as designed, automatically rebalancing the mining target to maintain the 10-minute block cadence.
The concern is directional. If AI contracts lock in mining capacity for 5–15 years, as current deals suggest, the hashrate that exits is not coming back — regardless of Bitcoin's price. CoinShares estimates hashrate could recover to 1.8 ZH/s by year-end, conditional on bitcoin recovering toward $100,000. At $76,000, that projection appears optimistic.
The more fundamental question is whether Bitcoin's fee market — currently generating approximately $55 million annualized on-chain fee revenue according to webthreepedia's foundational economic value research — can eventually replace the block subsidy as the primary incentive for miners. The current data suggests it cannot, not at these price levels and not at these transaction volumes.
The AI pivot introduces a secondary concern: geographic and corporate concentration. Publicly listed U.S. miners have accounted for over 40% of global hashrate. As these entities redirect capacity to AI, the remaining hashrate concentrates among private operators, many in regions with limited regulatory transparency.
ASIC production remains dominated by three firms: Bitmain, MicroBT, and Canaan. Supply chain disruptions at the hardware level cascade directly into mining capacity constraints. The top two mining pools already control more than 50% of block production; the top six account for 80–90%.
Paradoxically, some analysts argue the exodus of U.S. listed miners from bitcoin could improve geographic decentralization, as hashrate redistributes to lower-cost jurisdictions. Paraguay and Ethiopia have entered the global top 10 mining countries, driven by HIVE's 300 MW Paraguay operation and Bitdeer's 40 MW Ethiopia facility.
CoinShares' Q1 2026 report projects:
Bloomberg reported in April 2026 that AI revenue was on track to surpass bitcoin mining revenue for leading crypto mining companies, validating the CoinShares thesis.
The bitcoin mining sector is undergoing a structural transformation, not a cyclical adjustment. The combination of post-halving economics, depressed hash prices, and the availability of contracted AI revenue at 3–5x mining returns has created a one-way door for many operators. Companies that control energized power infrastructure are rationally redirecting that capacity toward its highest-value use. That use is no longer bitcoin mining.
For Bitcoin the network, the short-term impact is manageable. The difficulty adjustment mechanism is functioning as designed, and absolute hashrate levels remain high by historical standards. The long-term concern is that AI contracts are multi-year commitments that structurally remove capacity from the mining market, regardless of future BTC price movements. Bitcoin's security model ultimately depends on miners finding it economically rational to allocate energy to hashing. At current fee levels and block subsidy schedules, the competition from AI workloads represents a persistent drain on that economic rationality.
The sector's transformation also highlights a finding consistent with broader blockchain economic analysis: bitcoin mining has long operated as a subsidy-dependent activity, with approximately $18 billion in annual issuance subsidizing a network that generates $55 million in organic fee revenue. The AI pivot is, in one reading, the market pricing that subsidy gap in real time.