Bitcoin mining at scale is dominated by a cohort of publicly listed companies that operate large data centers, disclose their hashrate monthly, and finance hardware through equity and debt markets. Their disclosures make the industry unusually legible, and their reported figures need reading carefully, because companies use different definitions of capacity and trackers reconcile them differently.
Reading the figures before reading the ranking
Three separate sources consulted on 8 September 2026 gave materially different hashrate figures for the same companies. One aggregator listed a leading miner at 76.7 exahashes per second while industry reporting for the same period put the figure at 63.0. Another put a large operator between 57 and 72 exahashes per second depending on the source and the month.
The discrepancies are definitional rather than dishonest. A company may report peak hashrate, monthly average hashrate, energized capacity, or self-mined hashrate excluding hosted customers, and these differ by wide margins. Any ranking of miners is therefore directional, and a table presenting single figures to one decimal place is conveying more precision than the underlying disclosures support.
Who the large operators are
The listed cohort includes MARA Holdings, CleanSpark, Riot Platforms, Core Scientific, Bitdeer, IREN, Cipher Mining, TeraWulf, Hut 8, Bitfarms, HIVE Digital and several newer entrants. Between them they operate sites concentrated in the United States, with additional capacity in Canada, Scandinavia, Bhutan and parts of the Middle East.
Their disclosures give a useful picture of what running at scale involves. One operator reported mining 593 BTC in August 2026, holding 13,703 BTC on its balance sheet at the end of that month, running peak fleet efficiency of 16.07 joules per terahash, and having 1.8 gigawatts of contracted power of which 808 megawatts were in use.
That last ratio is the industry's current preoccupation. Contracted power well ahead of power actually consumed is the signature of a company holding capacity for something other than mining.
The pivot to AI and high-performance computing
The defining industry development of 2025 and 2026 has been mining companies converting data center capacity to AI and high-performance computing hosting. The logic is that a company holding grid connections, substations and cooling infrastructure owns the scarce input that AI compute demand is short of, and that hosting contracts pay contracted revenue rather than exposing the business to hashprice.
The scale of the shift is visible in revenue mix. One large operator reported colocation revenue of $136.7 million against $27.5 million from mining in the second quarter of 2026, colocation having risen to 83 percent of sales from 67 percent the previous quarter. Another reported high-performance computing lease revenue of $31.9 million against $12.8 million from mining in the same period.
Deal sizes have followed. A 15-year lease announced in July 2026 for the second phase of one company's Texas campus was reported at $9.8 billion, adding 352 megawatts and taking the tenant's footprint at the site to 704 megawatts. Another operator announced $2.8 billion in new multi-year AI cloud contracts the same month.
What that has done to hashrate
Converting capacity means decommissioning miners. Industry reporting put the tracked public cohort's combined hashrate at 319.0 exahashes per second in the second quarter of 2026, down from 368.3 in the fourth quarter of 2025, a decline of 13.4 percent at a time when the network as a whole fell by about 10.6 percent.
Excluding one company that expanded rather than contracted, the rest of the cohort fell 21.2 percent, from 324.6 to 255.9 exahashes per second. Several operators have decommissioned their mining fleets entirely to redirect sites to data center construction.
| Bitcoin mining | AI and HPC hosting | |
|---|---|---|
| Revenue | Variable with hashprice and difficulty | Contracted, often multi-year |
| Capital intensity | Machines depreciate fast | Buildings and cooling last longer |
| Counterparty | The protocol | A named tenant with credit risk |
| Flexibility | Can curtail within seconds | Contractual uptime obligations |
The tradeoff explains why not everyone has moved. Mining revenue is volatile but the customer never defaults and load can be curtailed instantly, which is what makes miners valuable to grid operators. Hosting revenue is contracted but introduces tenant credit risk and uptime obligations that remove the flexibility grids pay for.
Consolidation
The sector's largest attempted transaction did not complete. An all-stock acquisition of one large miner by an AI cloud company, valued at roughly $9 billion, was rejected by the target's shareholders on 30 October 2025 and terminated. The target subsequently grew its own hosting business to the majority of its revenue independently.
Financial pressure on pure-play mining has been visible elsewhere. One large operator reported a $611 million loss in the second quarter of 2026 with revenue down 27 percent, and described its strategy as a digital infrastructure pivot.
How to read a miner's disclosure
Monthly operational updates from listed miners follow a common format, and a few figures carry most of the information. Hashrate tells you scale but needs its definition checked. Coins mined in the month, divided by average hashrate, tells you whether the fleet is actually running or sitting idle behind a headline capacity number.
Fleet efficiency in joules per terahash determines the electricity bill per unit of output and is the single best predictor of who survives a hashprice decline. Coins held on the balance sheet indicate whether the company sells production to fund operations or accumulates it, which changes how its share price behaves relative to the coin. Contracted power against power consumed shows how much capacity is being held for future use.
Public miners and pool tables are different measurements
A common error is comparing a list of mining companies against a list of mining pools. Pool tables measure where hashrate is directed for the purpose of finding blocks. Company tables measure who owns and operates machines. A company can run its own pool, point its hashrate at someone else's, or host machines it does not own.
Both tables are useful and they answer different questions. For network concentration, the pool table is the relevant one. For industry structure and who is financing new capacity, the company disclosures are.