Public Bitcoin miners shed an estimated 75 EH/s of realized hashrate during the first half of 2026 as weaker mining economics and the race to build AI infrastructure changed how operators used their power. TheEnergyMag estimates that amount of mining capacity represents about $1.5 billion of ASIC hardware at an assumed acquisition price of $20 per TH/s.
The more revealing number is 75 EH/s itself. Against Bitcoin’s current seven-day network estimate of roughly 997 EH/s, the capacity public miners shed equals about 7.5% of today’s entire Bitcoin network. That is large enough to show why the AI transition is more than a corporate strategy story: it is changing where industrial-scale SHA-256 compute lives.
75 EH/s Is About 7.5% of Today’s Network
BitcoinVersus tracked the network near 997.03 EH/s on October 9. Divide the estimated 75 EH/s shed by public miners by that figure and the result is about 7.52%.
If that 75 EH/s were suddenly added back to the current network, the simple arithmetic would put total hashrate near 1.072 ZH/s. Bitcoin’s difficulty adjustment would eventually respond if the extra compute remained online, so that should not be read as free additional miner revenue. It does show the scale of compute public miners have already redirected, retired, sold, or otherwise removed from realized production.
The network did not lose 7.5% today because the 75 EH/s reduction occurred earlier in 2026 and other miners expanded afterward. Private operators, sovereign projects, newer hardware, and surviving public fleets helped refill the gap. BitcoinVersus recently covered how private and sovereign miners gained network share while public operators changed strategy.
The Missing Hashrate Represents More Than 1 GW of ASIC Load
Hashrate can also be translated into approximate electrical load. At an average fleet efficiency of 15 J/TH, 75 EH/s would require about 1.125 GW at the ASIC level. At 17.5 J/TH, the same hashrate requires about 1.31 GW. At 20 J/TH, it reaches about 1.5 GW.
Those figures exclude pumps, fans, transformers, networking, lighting and other site loads, so total facility demand would be higher. They also explain why the AI opportunity is centered on the power connection rather than the mining machine. The ASIC itself cannot become a GPU server, but a site with a substation, utility agreement, land, fiber access and experienced operations team may have infrastructure that can support another compute workload after substantial redesign.
That distinction is central to the industry’s current transition. BitcoinVersus previously examined why miners are shifting megawatts from ASICs toward AI. The new 75 EH/s estimate puts a compute number on the physical side of that shift.
At Today’s Hashprice, 75 EH/s Would Gross Nearly $3 Million a Day
There is also an opportunity cost on the mining side. At the roughly $39.14 per PH/s/day spot hashprice BitcoinVersus tracked today, 75 EH/s—or 75,000 PH/s—would correspond to about $2.94 million per day of gross mining revenue if it were earning the current hashprice.
Annualized without changing any variable, that is roughly $1.07 billion of gross revenue. That is not a profit estimate and it is not what the removed equipment necessarily would have earned. If 75 EH/s re-entered the network permanently, difficulty would rise and unit hashprice would fall. Electricity, pool fees, maintenance and site overhead would also consume a large part of gross revenue.
Still, the calculation shows the hurdle AI projects must clear. Operators are not replacing a worthless workload. They are choosing to redirect scarce power away from an activity that, at current conditions, can still generate billions of dollars of annual industry revenue.
The $1.5 Billion Figure Is an Estimate, Not a Reported Loss
TheEnergyMag’s $1.5 billion figure should be interpreted carefully. It multiplies 75 EH/s by an assumed $20 per TH/s hardware acquisition value. It is an equivalent hardware-investment estimate, not a disclosed cash loss, sale price, or impairment charge.
The same review separately found roughly $1.1 billion of asset impairments and held-for-sale markdowns among 12 tracked companies during the first half of 2026. IREN and Core Scientific represented almost 89% of that total. Those accounting charges overlap conceptually with the hardware transition but are not the same measurement as the $1.5 billion hashrate-equivalent estimate.
That difference matters because mining equipment does not disappear when a site converts. Machines may be sold, moved to another facility, written down, stored, or operated elsewhere. The real economic loss depends on what the owner paid, how much Bitcoin the machines produced, their resale value, remaining useful life, and what replaces them.
Newer ASICs Make the Decision Harder
The opportunity cost is not equal across every machine. A high-efficiency ASIC can remain valuable at hashprices that force older miners offline. BitcoinVersus just showed how a five-joule efficiency difference between two Canaan machines changed which one stayed online during weak economics.
That makes converting a site full of relatively new hardware different from retiring a fleet already near its electrical break-even point. The better the ASICs and the cheaper the power, the more AI revenue must justify the lost mining opportunity.
TheEnergyMag gives a concrete example at Cipher’s Black Pearl facility. The site began mining in mid-2025, yet by year-end Cipher had recorded a $96.1 million markdown on Black Pearl mining machines as the location moved toward high-performance computing. The machines generated about $57.9 million of 2025 revenue before the transition. That does not establish the project’s lifetime return, but it shows how quickly the economic use of a newly built mining site can change.
The Power Connection Is Becoming More Valuable Than the ASIC
The 75 EH/s estimate points to a broader change in how mining companies value their assets. The machine used to be the center of the expansion story: buy newer ASICs, install more exahashes, and defend network share. In the AI era, energized land, substations, transformers, transmission access, cooling rights, fiber, and interconnection position can command more strategic value than the SHA-256 hardware sitting behind them.
That does not mean every Bitcoin mine can become an AI data center. AI facilities need higher network bandwidth, more redundancy, different cooling architectures, rack-level redesign, stricter uptime requirements and often major additional capital. Galaxy Research has emphasized that only mining sites with the right power, acreage, fiber, water or cooling access, permitting and skilled labor are strong conversion candidates.
Watch Why Miners Are Moving Toward AI
The video below discusses why the AI pivot is mostly available to large operators with power-ready infrastructure—and why smaller miners cannot simply replace an ASIC with a GPU server.
What 75 EH/s Tells Us
- 75 EH/s equals about 7.5% of Bitcoin’s current seven-day network hashrate estimate.
- At 15–20 J/TH, that hashrate represents roughly 1.1–1.5 GW of ASIC-level electrical load.
- At a $39.14 hashprice, 75 EH/s corresponds to about $2.94 million per day of gross mining revenue before costs and difficulty response.
- The $1.5 billion hardware figure is an equivalent-value estimate using $20/TH, not a reported realized loss.
- The strategic asset being preserved through many AI conversions is the power-ready site—not the ASIC itself.
The biggest inference is that Bitcoin mining’s AI transition is not mainly about abandoning Bitcoin for another technology. It is about deciding what workload deserves a scarce megawatt. Public miners already removed enough realized hashrate to equal roughly one-thirteenth of today’s network, yet Bitcoin’s total hashrate recovered toward 1 ZH/s anyway. The network replaced the compute. The companies are trying to replace the economics.
Leave a Reply