Bitcoin's network hashrate peaked near 1.3 zettahashes per second in late 2025, then fell 22-24% over the following months. As of September 7, 2026, the seven-day average sits near 934 EH/s, according to CoinWarz. Hashprice — the dollar revenue per petahash per day — recorded its three lowest mon...
"We're living, I think, through the first bear market in hash rate that we've ever seen in Bitcoin history." — Rapha Zagury, CEO, Twenty One Capital
Bitcoin's network hashrate peaked near 1.3 zettahashes per second in late 2025, then fell 22-24% over the following months. As of September 7, 2026, the seven-day average sits near 934 EH/s, according to CoinWarz. Hashprice — the dollar revenue per petahash per day — recorded its three lowest monthly averages in history during H1 2026: $30.37 in June, $31.21 in July, and $31.27 in March.
The decline is not cyclical. It is structural. Listed Bitcoin miners have signed over $70 billion in aggregate AI and high-performance computing (HPC) contracts since 2025, according to Noah Intelligence. Yet actual AI revenue in H1 2026 totaled $341.2 million against $5.11 billion in capital expenditures. The industry is mid-transition: miners are spending billions to convert power infrastructure from SHA-256 hashing to GPU-based AI workloads, and the economics make the case clear. According to HIVE Digital Technologies, 10 MW of Nvidia H100 GPUs produces revenue comparable to 100 MW of Bitcoin mining.
This report examines the structural forces reshaping Bitcoin mining into a bifurcated industry — one half optimizing for AI compute margins, the other migrating to low-cost hydroelectric jurisdictions in the Global South.
Twenty One Capital CEO Rapha Zagury identified Bitcoin's current condition at Bitcoin Asia 2026 in Hong Kong: the network is in its first-ever hashrate bear market. Unlike previous hashrate declines — China's 2021 mining ban caused a 50% drop that recovered in five months — the current drawdown is gradual, sustained, and driven by rational capital allocation rather than regulatory shock.
The network touched 1,001 EH/s on September 4, 2026, per CoinWarz, but this was a brief spike. The sustained seven-day average remains below the 1 ZH/s threshold that was first breached in January 2026. The peak-to-trough decline of 22-24% represents the longest continuous drawdown in Bitcoin's mining history that was not triggered by a single external event.
According to a September 2, 2026, analysis by Cryptonomist, the decline is structural rather than cyclical. Mining companies are not shutting down due to insolvency. They are reallocating power capacity, land, and cooling infrastructure to AI workloads that generate higher and more predictable returns per megawatt.
The April 2024 halving cut block rewards from 6.25 BTC to 3.125 BTC. Two years later, the margin squeeze is fully visible:
The economics create a binary outcome. Operators with sub-$0.05/kWh power and latest-generation ASICs remain profitable. Everyone else faces a choice: upgrade hardware, find cheaper power, pivot to AI, or shut down.
The aggregate numbers tell a story of ambition outpacing execution. According to Noah Intelligence, listed miners announced $70 billion in AI/HPC contracts through H1 2026. Actual AI revenue in the same period: $341.2 million. CoinShares projects AI could represent 70% of listed miner revenue by year-end 2026, up from approximately 30% in early 2026.
Capital expenditure data from Cointelegraph shows miners spent $5.11 billion on capital assets in H1 2026 — predominantly on AI infrastructure conversion. The gap between contract value and realized revenue reflects construction timelines: most contracts are 12- to 20-year agreements with revenue recognition beginning 12-18 months after signing.
The economic rationale is straightforward. Iris Energy (IREN) reported that a small Nvidia GPU deployment generated 3-4x the economic return versus self-mining on equivalent power capacity. This differential is driving an industry-wide reallocation.
Core Scientific (CORZ): The most advanced in the transition. Over $10 billion in contracted revenue anchored by a 12-year agreement with CoreWeave across approximately 590 MW of critical IT load at six sites. Target cash gross profit margin: 80-85%. The company sought a $3.3 billion bond offering in April 2026 to fund further AI data center construction. Total leasable power pipeline: approximately 3.0 GW.
TeraWulf (WULF): Signed a 20-year lease with Anthropic for 401 MW of AI data center capacity at its Justified Data campus in Hawesville, Kentucky, starting 2027. Expected revenue: $19 billion over the initial term — exceeding TeraWulf's ~$12 billion market capitalization at signing. Q1 2026 revenue: $34 million, with 60% from AI compute (up 117% quarter-over-quarter). Bitcoin mining revenue fell 50%. The company sold its 50.1% stake in a Texas mining joint venture for ~$530 million and liquidated nearly all BTC holdings.
IREN (formerly Iris Energy): Completed a $3 billion convertible notes deal in May 2026 to fund its AI transition. Signed a $9.7 billion, five-year agreement with Microsoft for 200 MW of GPU cloud capacity. Q4 fiscal 2026 revenue: $137.2 million, with AI cloud services ($70.5M) surpassing Bitcoin mining ($66.7M) for the first time. Full fiscal year 2026 revenue: $707 million.
Cipher Mining (CIFR): Secured a 15-year, 300 MW direct lease with AWS, expected to generate $5.5 billion in revenue. Subsidiary Black Pearl Compute announced a $2 billion senior secured note offering in February 2026.
Marathon Digital (MARA): Pursuing a more gradual transition. Q2 2026 revenue: $174.9 million, down 26.67% year-over-year. Formed a strategic joint venture with Starwood Digital Ventures for AI/HPC infrastructure. Targeting at least two AI/HPC leases by year-end 2026. Needham lowered its 2026 revenue estimate to $736 million from $801 million.
Hyperscale Data (GPUS): Shut down all bitcoin miners at its Michigan site on September 1, 2026, freeing 20 MW for a California-based neocloud provider under a $1.2 billion, 20-year master services agreement. BTC holdings dropped 79%, from 1,006 BTC to ~215 BTC. Stock hit an all-time low of $0.22, down over 76% in 2026.
As North American miners convert facilities to AI, Bitcoin mining is migrating to jurisdictions with surplus hydroelectric power and low marginal electricity costs.
Paraguay: Holds 4.0% of global hashrate as of January 2026, per Hashrate Index. Powered by surplus capacity from the Itaipú Dam, with electricity costs as low as $0.033/kWh. HIVE Digital is expanding to 400 MW capacity, targeting 35 EH/s by year-end (up from 24 EH/s in late 2025). A 100 MW expansion began construction in early 2026, with commissioning expected in Q3.
Ethiopia: Holds 2.6% of global hashrate. Approximately 23 mining operations consume ~600 MW, powered by surplus hydroelectric capacity from the Grand Ethiopian Renaissance Dam (inaugurated September 2025). Production cost: approximately $16,000 per BTC — among the lowest globally. BitX signed an MOU with Ethiopia Mining Farm in May 2025 for accelerator deployment.
Broader trend: The lowest-cost mining jurisdictions are increasingly in the Global South, where stranded hydroelectric power creates natural economic advantages. This geographic redistribution has implications for Bitcoin's decentralization thesis and for the energy profiles of mining operations worldwide.
According to the Cambridge Centre for Alternative Finance (CCAF), which surveyed 49 mining companies across 16 jurisdictions (representing 48% of global hashrate), Bitcoin mining's energy mix in 2026 breaks down as follows:
| Source | Share | |--------|-------| | Hydropower | 23.4% | | Wind | 15.4% | | Solar | 3.2% | | Other renewables | 0.6% | | Nuclear | 9.8% | | Total sustainable | 52.4% | | Natural gas | 38.2% | | Coal | 8.9% | | Other fossil | 0.5% |
The industry that once drew over a third of its electricity from coal (37.6% in 2022) now sources more than half from sustainable generation. The AI pivot reinforces this trend: AI data center contracts from hyperscalers like Microsoft, AWS, and Anthropic typically require renewable energy commitments, pushing converted facilities further toward clean power.
The hashrate decline raises a question that the industry has rarely confronted: what happens to Bitcoin network security when mining infrastructure becomes more valuable as AI compute?
At ~934 EH/s, the network remains orders of magnitude more secure than any other proof-of-work chain. A 51% attack would require approximately 467 EH/s of sustained hashrate — an economically implausible proposition at current hardware costs.
However, the structural trend is clear. Capital is flowing away from SHA-256 ASICs and toward GPUs. If AI revenue premiums persist — and the 3-4x return differential reported by IREN suggests they will — the industry may reach an equilibrium where only operators in the lowest-cost power markets continue mining Bitcoin, while higher-cost jurisdictions permanently convert to AI.
The countervailing force is Bitcoin's difficulty adjustment mechanism. As hashrate declines, difficulty adjusts downward (a 19% decrease was recorded in early 2026), reducing the cost of mining for remaining participants and re-establishing profitability. This self-correcting mechanism prevents a permanent security spiral but does not prevent geographic concentration.
The Bitcoin mining industry is undergoing a structural transformation with no historical precedent. The 2024 halving compressed margins. The AI infrastructure shortage created a higher-value use for miners' core assets: power contracts, substations, cooling systems, and operating expertise. The result is an industry bifurcating into two distinct segments — AI infrastructure companies with mining heritage, and lean mining operations in low-cost hydroelectric markets.
The $70 billion in announced contracts versus $341 million in realized revenue represents the industry's current position: committed but not yet converted. Over the next 12-18 months, as facilities complete construction and revenue recognition begins, the financial profiles of these companies will shift from Bitcoin-dependent miners to diversified compute infrastructure providers.
For Bitcoin's network, the implications are nuanced. Security remains robust at current hashrate levels. The difficulty adjustment ensures mining remains viable for efficient operators. But the era of hashrate growth as a proxy for industry health is over. The miners that once competed to secure Bitcoin's blockchain are now competing to host the AI models that will define the next decade of computing.