Bitcoin mining companies have signed an estimated $150 billion in cumulative AI data center contracts, according to Bernstein, covering more than 7.5 gigawatts of electrical capacity across multi-year terms. The pivot is structural, not speculative: MARA Holdings posted a 27% year-over-year reven...
"Mining revenue is projected to plummet from around 85% of total revenue in early 2025 to less than 20% by the end of 2026 for companies that have secured AI contracts." — CoinShares, Q1 2026 Bitcoin Mining Report
Bitcoin mining companies have signed an estimated $150 billion in cumulative AI data center contracts, according to Bernstein, covering more than 7.5 gigawatts of electrical capacity across multi-year terms. The pivot is structural, not speculative: MARA Holdings posted a 27% year-over-year revenue decline in Q2 2026, CleanSpark fell 30.5%, and the weighted average cash cost to produce one bitcoin reached $79,995 in Q4 2025 — roughly $16,000 above BTC's current trading price of approximately $64,000.
Meanwhile, the network's hashrate concentration has reached levels that challenge Bitcoin's foundational decentralization premise. The Nakamoto coefficient — the minimum number of pools needed to exceed 50% of hashrate — has fallen to 3, according to D-Central's H1 2026 report. Foundry USA alone produces approximately 27% of all blocks. SBI Crypto, a subsidiary of Japanese financial giant SBI Holdings, shut down its mining pool on July 31, 2026, after five years of operation, removing 20.9 EH/s (2.2% of network hashrate) in a single day.
The mining industry is splitting into two distinct businesses: AI infrastructure landlords generating 80-90% operating margins on long-term leases, and pure-play BTC miners operating at or below breakeven with a hashprice stuck near five-year lows. The question is not whether this transformation continues, but how much of the original mining business survives it.
The Bitcoin network operates at approximately 956-1,003 EH/s as of August 7-11, 2026 (methodology-dependent), according to CoinWarz and KuCoin data. Network difficulty stands at 126.23 trillion, near all-time highs, with the next retarget estimated at 128.4 T (+0.72%) around August 22.
The network briefly crossed the 1 ZettaHash/s (1,000 EH/s) mark in 2025, then went through an approximately 25% miner-capitulation purge around the turn of the year. It has since stabilized in the 950-1,000 EH/s range. Hashprice — the revenue per petahash per day — sits at $29-33/PH/s/day, a level D-Central characterizes as "reminiscent of 2020 after the COVID crash."
Publicly traded miners tracked by BitcoinMiningStock.io operate 404.5 EH/s combined, accounting for roughly 44% of the network. The remaining 56% is distributed across private operations, sovereign miners, and unidentified pools.
Bitcoin's mining pool market has consolidated to a degree that warrants structural concern. As of August 10, 2026, the top four pools control the following shares:
| Pool | 7-Day Hashrate Share | Operator | |------|---------------------|----------| | Foundry USA | 24.3% | Digital Currency Group | | AntPool | 17.5% | Bitmain | | ViaBTC | ~13% | Independent (CIS exposure) | | F2Pool | ~12% | Independent |
These four pools collectively command approximately 67-70% of all blocks mined. The top six pools produce 95-99% of blocks, according to D-Central's pool centralization tracker.
The Nakamoto coefficient of 3 means only three pools are needed to coordinate a theoretical majority attack. This is not an abstract concern — it is a measurable deviation from the design assumptions of a permissionless proof-of-work system.
Foundry USA, backed by Digital Currency Group, primarily serves institutional clients with custom terms, dedicated support, and preferential fee structures. This creates a two-tier market structure where large operators receive different service levels than smaller miners, according to D-Central's H1 2026 mining report.
ViaBTC has faced regulatory scrutiny affecting miners tied to Russia and other CIS countries, including account restrictions, sudden KYC demands, and temporary fund freezes, according to the same report.
The post-halving economic reality is stark. The April 2024 halving reduced the block subsidy from 6.25 BTC to 3.125 BTC. Combined with difficulty remaining near all-time highs and BTC trading in the mid-$60,000s, the math has turned negative for a significant portion of the fleet.
Key economic data points:
CoinDesk estimated per-BTC losses of approximately $19,000 at the prevailing hashprice and difficulty level, explaining why miners are net sellers of BTC in 2026 to fund operational transitions.
The migration from BTC mining to AI/HPC (high-performance computing) infrastructure has moved from strategy to execution. According to Bernstein, Bitcoin miners have signed contracts worth an estimated $150 billion in aggregate, covering more than 7.5 GW of electrical capacity across terms spanning 10-20 years. CoinShares' more conservative estimate puts the figure at over $70 billion.
Major contract announcements (2026):
| Company | Counterparty | Capacity | Contract Value | Term | |---------|-------------|----------|---------------|------| | Core Scientific | CoreWeave | 590 MW | $10.2B | 12 years | | Core Scientific | AMD | 529+ MW (up to 2.5 GW) | $14B+ | 15 years | | TeraWulf | Anthropic | 401 MW | $19B | 20 years | | IREN | Microsoft | ~60 MW | $1.94B annualized | 5 years | | IREN | NVIDIA | ~60 MW | $3.4B | 5 years | | Hut 8 | Undisclosed | 1 GW (Beacon Point) | $9.8B | 15 years | | Riot Platforms | Undisclosed ("leading frontier AI lab") | 191 MW | $9.1B | 20 years |
HIVE Digital Technologies has estimated that 10 MW of NVIDIA H100 GPUs can produce revenue comparable to 100 MW of Bitcoin mining — a 10:1 revenue density advantage.
These contracts carry 80-90% project-level EBITDA margins, according to IREN's disclosures, compared to the razor-thin or negative margins on BTC mining at current hashprice levels.
The economic logic is unambiguous: miners own the three assets AI companies need most — grid interconnections, land with utility agreements, and operational teams experienced in managing high-density compute at scale. The pivot monetizes infrastructure that was increasingly unprofitable for its original purpose.
Q2 2026 earnings for public miners reflect the industry's bifurcated reality:
MARA Holdings (Marathon Digital):
CleanSpark:
Riot Platforms:
The revenue declines at MARA and CleanSpark — 27% and 30.5% respectively — demonstrate the direct impact of halved block subsidies and elevated difficulty on pure-play mining revenue. Riot's 14% revenue increase was driven entirely by its engineering and data center segments, not mining.
Combined net losses across the three companies totaled approximately $1.09 billion in a single quarter. This is not a temporary earnings miss — it reflects the structural reality that mining BTC at $64,000 with a $79,995 average production cost produces losses.
SBI Crypto's July 31, 2026 shutdown illustrates the rational exit calculus for operators without a viable AI pivot path.
Timeline:
SBI Holdings did not disclose a reason for the shutdown, according to CoinDesk. However, the parent company has been expanding its digital-asset business through other channels, recently pursuing a $289 million stake in Japanese crypto exchange Bitbank, spanning trading, custody, and payments.
The exit aligns with a pattern: operators who lack access to cheap stranded power (sub-$0.04/kWh), who cannot access AI contract revenue, or who face unfavorable regulatory environments are finding that the mining operation has negative expected value. SBI's pivot from mining to exchange equity is a capital reallocation decision, not a crypto exit.
1. Hashrate concentration creates censorship vectors. With a Nakamoto coefficient of 3, coordinated action by just three pool operators could theoretically censor transactions. This does not require malicious intent — regulatory pressure on identifiable pool operators could produce the same outcome.
2. The AI pivot creates execution risk. The $150 billion contract pipeline is measured in nominal revenue over 10-20 year terms. Contracts of this duration carry counterparty risk, technology obsolescence risk (GPU generations turn over every 18-24 months), and energy price risk. Miners are trading one form of commodity exposure (BTC price) for another (AI compute demand).
3. Mining may become a loss leader. For companies like Riot that derive increasing revenue from engineering and data center segments, BTC mining could persist as a portfolio allocation or optionality play rather than a primary business. At sub-$30 hashprice, mining exists only for operators with structural cost advantages or for those treating it as a BTC accumulation strategy.
4. Geographic concentration compounds pool concentration. The United States dominates publicly listed mining capacity (404.5 EH/s of tracked public hashrate is heavily US-weighted), creating jurisdictional concentration atop pool concentration. SBI's exit from Japan removes one of the few remaining non-US institutional pool operators.
The Bitcoin mining industry in August 2026 is undergoing a structural transformation that has no precedent in its 17-year history. The economics of proof-of-work mining — post-halving, at record difficulty, with hashprice at five-year lows — have made pure BTC production unprofitable for average-cost operators. The rational response, and the one the market is executing, is capital reallocation toward AI infrastructure, where the same physical assets generate 10x the revenue density.
This creates a paradox for Bitcoin's security model. The network requires miners to operate; the most capable mining companies are reallocating their best assets to AI. What remains is an increasingly concentrated pool structure where three operators control majority hashrate, and a cost structure where only the lowest-cost quartile of miners can operate profitably.
The $150 billion AI contract pipeline validates the infrastructure assets these companies built. It does not validate the sustainability of the mining business that created those assets.