Public Bitcoin miners reported Q2 2026 earnings that confirm an accelerating structural shift away from cryptocurrency mining toward artificial intelligence data center hosting. MARA Holdings, Riot Platforms, CleanSpark, and Core Scientific collectively posted combined net losses exceeding $2.2 b...
"You get a lot more money per electron if you're doing it for AI than for Bitcoin mining." — Fred Thiel, CEO, MARA Holdings
Public Bitcoin miners reported Q2 2026 earnings that confirm an accelerating structural shift away from cryptocurrency mining toward artificial intelligence data center hosting. MARA Holdings, Riot Platforms, CleanSpark, and Core Scientific collectively posted combined net losses exceeding $2.2 billion for the quarter, driven by unrealized fair-value losses on Bitcoin holdings and write-downs of mining equipment. Mining revenue fell double digits year-over-year at MARA (−27%) and CleanSpark (−25%), while Core Scientific's total revenue doubled to $164.2 million — with 83% now coming from AI colocation, not mining.
The migration is structural, not cyclical. Hash price — the revenue a miner earns per unit of computational power — fell to $23.9 per PH/s per day in Q1 2026, its lowest level since 2018, according to CoinShares. At that rate, CoinShares estimates 15–20% of the global mining fleet is operating below breakeven. Public miners have responded by signing more than $70 billion in cumulative AI and high-performance computing (HPC) contracts, and CoinShares projects that 70% of listed miner revenue will derive from AI workloads by year-end 2026.
The implications extend beyond individual company strategy. As miners redirect electrical capacity from SHA-256 hashing to GPU-based AI inference and training, the Bitcoin network faces a potential long-term question about hashrate growth and security budgets — even as the network's difficulty sits at 127.48 trillion and hashrate hovers near 878 EH/s.
The four largest publicly traded Bitcoin miners by market capitalization reported Q2 2026 results that reveal a sector in transition:
| Company | Q2 Revenue | YoY Change | Net Loss | BTC Mined | AI/HPC Revenue Share | |---------|-----------|------------|----------|-----------|---------------------| | MARA Holdings | $174.9M | −27% | −$611.3M | 2,422 | Minimal | | Riot Platforms | $174.2M | +14% | −$237.2M | N/A | 21% ($37.3M) | | Core Scientific | $164.2M | +109% | −$1.16B | N/A | 83% ($136.7M) | | CleanSpark | $136.4M | −25% | −$378M | 1,799 | Early stage |
The revenue divergence is notable. MARA and CleanSpark, which remain primarily dependent on Bitcoin mining, posted revenue declines. Riot, which has begun diversifying, grew revenue 14% year-over-year as its data center and engineering segment generated $37.3 million, up from $10.6 million in Q2 2025. Core Scientific, the furthest along in its AI pivot, doubled revenue, with AI colocation contributing $136.7 million — up from $10.6 million in the same quarter a year prior.
Net losses across all four companies were inflated by GAAP fair-value accounting on Bitcoin holdings. MARA reported $343 million in unrealized losses on digital assets. Core Scientific's $1.16 billion loss included a $1.05 billion increase in the fair value of warrant liabilities. These are accounting artifacts, not operating cash burns, but they underscore the volatility penalty embedded in the mining-heavy business model.
The April 2024 halving reduced the block subsidy from 6.25 BTC to 3.125 BTC. Two years later, the full weight of that revenue cut is visible in operating data.
Hash price collapse. Hash price — the dollar revenue per petahash per second per day — fell roughly 80% from its pre-halving peak of approximately $120/PH/s/day to $23.9/PH/s/day by Q1 2026, according to CoinShares' Q1 2026 mining report. This represents the lowest reading since 2018.
Production costs exceed market price. CoinShares reports the weighted average all-in cost among publicly listed miners reached approximately $80,000 per Bitcoin in Q4 2025. With Bitcoin trading near $77,700 in August 2026, many miners are producing coins at a loss. The cost range is wide — a hydro-cooled Antminer S23 at 9.5 J/TH can produce at roughly $32,000 per coin, while an air-cooled S21 at 17.5 J/TH lands near $59,000. Legacy S19 fleets operating above 30 J/TH became cash-flow negative in most jurisdictions during the late-2025 hash price decline.
Electricity remains the dominant variable. Power costs represent 60–80% of total mining operating expenses. The profitability threshold sits at approximately $0.06–$0.08 per kilowatt-hour for modern hardware. The Cambridge Bitcoin Electricity Consumption Index estimated network power demand at 16.09 GW and annualized consumption at 141.02 TWh as of August 1, 2026 — equating to roughly 858,000 kWh per Bitcoin mined.
Difficulty persists. Mining difficulty reached 144.4 trillion in February 2026 — a 14.73% single-adjustment jump, the largest percentage increase since China's 2021 mining ban. Though difficulty has since moderated to 127.48 trillion as of mid-August, it continues to rise at an average rate of approximately 5% per month, compressing margins even as hardware efficiency improves.
Fleet obsolescence accelerates. New-generation ASICs operate at 15–20 J/TH, while the global fleet-wide average is 18.3 J/TH as of July 31, 2026. CoinShares estimates 15–20% of the global mining fleet is now unprofitable, creating forced seller dynamics among operators unable to upgrade hardware or secure cheaper power.
The economic calculus is straightforward. AI data center hosting generates higher revenue per megawatt than Bitcoin mining, with longer contract durations and more predictable cash flows.
Contract scale. Public miners have announced more than $70 billion in cumulative AI and HPC contracts. Riot Platforms signed a 20-year lease worth approximately $9.1 billion to host 191 MW of critical IT capacity for a frontier AI lab at its Rockdale, Texas facility. Core Scientific has signed 15-year agreements with AMD and Neocloud covering 529–530 MW, with AMD holding options for up to 2.5 GW. Core Scientific's total leased customer power capacity has reached 1.1 GW, representing over $24 billion in potential contracted revenue.
Revenue per megawatt. MARA CEO Fred Thiel has stated publicly that AI hosting yields more revenue per electron than Bitcoin mining. Core Scientific's Q2 results provide a partial benchmark: $136.7 million in colocation revenue against roughly 200–300 MW of deployed AI capacity suggests annualized revenue of approximately $1.8–$2.7 million per MW — meaningfully higher than the approximately $0.5–$1.0 million per MW that Bitcoin mining generates at current hash prices.
CoinShares projections. CoinShares' analysis projects 70% of listed miner revenues will come from AI workloads by end-2026. At current run rates, Core Scientific is already well past that threshold at 83%.
The transition is not costless. Converting mining facilities to GPU-ready AI infrastructure requires capital expenditure on power distribution, cooling systems, and network connectivity. Core Scientific recorded $266.5 million in impairments of mining-related assets during Q2, reflecting the write-down of equipment being retired or repurposed.
Core Scientific has executed the most complete pivot. AI colocation revenue grew from $10.6 million to $136.7 million year-over-year, representing 83% of Q2 revenue. Mining contributed just $27.5 million ($21.5 million self-mining, $6 million hosted mining). The company reported a consolidated gross margin of 43% and adjusted EBITDA of $41.1 million. Its partnership with AMD — covering potential capacity of 2.5 GW across 15-year terms — positions it as the largest pure-play AI infrastructure company to emerge from the mining sector.
Riot Platforms is mid-transition. Mining revenue was $113.7 million in Q2, still the dominant segment, but data center and engineering revenue tripled year-over-year to $37.3 million. The company ended Q2 with $1.2 billion in total liquidity ($666 million in Bitcoin, $549 million in cash) and deployed hashrate of 44.4 EH/s. The post-quarter $9.1 billion AI lease at Rockdale signals an aggressive acceleration of its HPC strategy.
MARA Holdings remains the most mining-dependent of the large public miners. Q2 revenue of $174.9 million was down 27% year-over-year, with an average selling price of $73,078 per Bitcoin mined. The $611.3 million net loss included $343 million in unrealized digital asset losses. MARA has signaled intent to diversify into AI infrastructure but has not yet reported material HPC revenue.
CleanSpark reported fiscal Q3 revenue (ended June 30) of $138 million, down 30.5% year-over-year. Bitcoin production remained stable at approximately 1,799 BTC for the quarter, with operational hashrate reaching 50.0 EH/s in April. Net loss was $378 million, including $263 million in non-cash mark-to-market adjustments. The company has entered hyperscaler lease discussions but has not yet closed AI hosting contracts.
The redirection of electrical capacity from mining to AI workloads has not yet materially affected Bitcoin's network hashrate. At 878 EH/s as of July 31, 2026, the hashrate remains near historic highs, having recovered from a weather-related dip to 826 EH/s earlier in the year.
However, the composition of who provides that hashrate is changing. As U.S.-listed miners redirect megawatts toward AI, private mining operations — particularly in regions with low-cost energy such as Kazakhstan, Russia, and parts of Africa — may absorb the freed capacity. The U.S., China, and Kazakhstan together account for over 75% of global Bitcoin mining electricity consumption.
The longer-term question is whether the AI pivot reduces the growth rate of hashrate, which has implications for network security. The Bitcoin security budget — total miner revenue from block subsidies and transaction fees — funds the cost of attacking the network. If hash price continues declining and AI offers superior returns on electrical infrastructure, the rate of new hashrate deployment could slow.
Transaction fees remain a marginal contributor, consistently below 1% of total block rewards in Q1 2026. Until fee revenue increases substantially — through higher on-chain transaction volume or adoption of fee-generating use cases — the security budget remains subsidy-dependent, and the subsidy halves again in 2028.
The Q2 2026 earnings cycle makes explicit what has been implicit for over a year: the publicly traded Bitcoin mining sector is becoming an AI infrastructure sector that happens to mine some Bitcoin. Core Scientific is the clearest example, generating more than five times as much revenue from AI colocation as from mining operations. Riot's $9.1 billion lease deal suggests the same trajectory.
This transition reflects rational capital allocation. At current hash prices and production costs, mining Bitcoin is a marginal business for most public operators. AI data center hosting offers higher revenue per megawatt, longer contract durations, and more predictable cash flows. The market has already internalized this: Nick Hansen, CEO of the Luxor mining pool, told DL News that "resisting the urge to transition to AI" would be Bitcoin miners' biggest challenge in 2026.
For the Bitcoin network, the near-term impact is limited. Hashrate remains near all-time highs and difficulty adjustments ensure the network self-corrects for capacity changes. The longer-term question — whether the AI pivot reduces the growth trajectory of hashrate and compresses the network's security budget ahead of the 2028 halving — remains open. The data available today does not support a definitive conclusion on that question.