Seven of the top ten miners by hashrate report AI or high-performance computing initiatives already generating revenue, with the other three planning to follow suit.
The shift pairs miners’ energized land and interconnections with contracted revenue from GPU customers, creating a second line of business that competes with running ASICs at full power.
AI partnerships redefine mining economics and investor focus
TeraWulf set the reference point after signing two 10-year hosting agreements with Fluidstack, totaling approximately 200 MW at Lake Mariner.
According to Barron’s, Google is backing a portion of Fluidstack’s lease obligations, up to approximately $ 1.8 billion, and has received warrants that could equate to roughly 8 percent of TeraWulf. The disclosed deal math implies roughly $1.85 million per MW per year of headline revenue over the term, which many miners now use as a benchmark when courting AI tenants.
Core Scientific expanded a 12-year relationship with CoreWeave for about 70 MW of additional HPC capacity, with operations targeted for the second half of 2025. Bitdeer continues to operate a commercial AI cloud based on NVIDIA DGX systems, while Iris Energy reports an AI cloud business running on H100 and H200 GPUs.
Others are building the real estate for the next wave. CleanSpark said on October 29 it secured 271 acres and about 285 MW of long-term power in Texas for what it calls a next-generation AI and HPC campus. Marathon agreed in August to acquire 64 percent of Exaion, an EDF subsidiary, to expand its global AI and HPC capabilities, with an option to increase its stake to 75 percent by 2027.
Riot has been assessing the conversion of approximately 600 MW at Corsicana for AI or HPC and has paused part of its mining expansion, resulting in a reduction of year-end 2025 hashrate guidance from 46.7 EH/s to 38.4 EH/s. Bitfarms has hired consultants to conduct a feasibility study and has been marketing its sites to AI clients.
Cipher Mining is reported to have a multi-year Fluidstack arrangement with a Google-linked lease commitment, although not all terms are disclosed in a single primary filing. Abu Dhabi’s Phoenix Group has signaled plans to scale alt=””>
The economic case rests on power and predictability.
Using today’s network context of about 1.08 to 1.10 ZH/s and 144 blocks per day with fees that have ranged from roughly 0.3 to 2.0 BTC per block, one MW of modern ASICs at about 17 J/TH translates to about 0.059 EH/s of hashrate.
That share of the network earns roughly $ 1.0 to $ 1.6 million per MW per year in gross mining revenue before power and opex, at a bitcoin price of nearly $104,000, according to CoinWarz data for price and hashrate. The midpoint of that range, around 1.2 to 1.3 million dollars, trails the 1.85 million dollars per MW per year implied by TeraWulf’s AI contracts.
Power price, capital expenditure (capex), and utilization determine margins in either model. Still, the contracted nature of AI hosting has become a key feature for equity investors seeking steadier cash flows rather than pure exposure to risk and fees.
Macro demand for>McKinsey charts show U.S.>ERCOT projects record peak demand over the next five years, with data centers a primary factor, as analyses indicate approximately 35 GW of peak>Reuters. Those numbers align with miners’ pitch that their grid ties, substations, and land banks are now scarce inputs for AI campuses, not just for exahash.
This realignment changes what matters inside the mining league table.
A miner that directs new megawatts toward AI may record lower headline hashrate growth than a pure-play operation. Yet, its enterprise value can improve through contracted revenue, power optionality, and longer-dated agreements.
Core Scientific’s additions with CoreWeave put a 12-year stamp on the model. CleanSpark’s 285 MW plan and Marathon’s Exaion purchase push miners toward owning and operating mixed-use campuses where GPUs, miners, and sometimes standard colocation can share infrastructure. Riot’s public evaluation of 600 MW at Corsicana demonstrates how quickly the mix can change when a site already has transformers, switchgear, water rights, and fiber infrastructure in place.
There are constraints. ERCOT interconnection timelines, gas turbine availability for new peakers, and transformer lead times all dictate how quickly high-density halls can be energized. GPU supply remains a swing factor as Blackwell and successor parts ramp and as hyperscalers allocate inventory to internal builds.
On the crypto side, any shift in fee regimes that materially lifts fees per block can close some of the per-MW revenue gap between mining and AI hosting. A move of about 0.5 BTC per block in sustained average fees is worth roughly $ 0.2 to $ 0.3 million per MW per year in miner gross revenue at current price levels, based on the simple share-of-network math above.
Investors are watching the composition of revenue, rather than just the exahash.
Contracted AI megawatts and dollars per MW per year are becoming the new disclosures to track. The $1.5 to $2.0 million per MW per year range is emerging as a practical benchmark for high-density hosting in the U.S., with TeraWulf’s disclosed figure serving as a current reference.
Utility capex plans and interconnection queue updates are now as relevant to miner outlooks as ASIC delivery schedules. As U.S. spot power tightens, miners with already energized land, permitted pads, and spare substations can monetize that optionality faster than greenfield entrants.
The international angle adds heft. Marathon’s move with Exaion ties a U.S. miner to an EDF affiliate inside the French power system, aligning GPU hosting with state-adjacent energy assets.
Phoenix Group’s plan to scale in the Gulf, while weighing a U.S. listing, puts sovereign power economics into the mix for AI infrastructure.
Those structures could pull more miners into joint ventures where utilities or energy investors anchor long-term contracts in return for capacity rights, priority interconnects, or equity stakes.
For crypto fundamentals, the pivot could slow the rate at which network hashrate expands through 2026 if material portions of new power are routed to GPUs instead of ASICs. The network will still add hash as new sites come online and as older fleets refresh, yet the slope can flatten relative to the last surge.
That would not stop capital from entering mining, as high bitcoin prices and fee spikes can still improve returns; however, it makes the hashrate leaderboard a weaker proxy for equity value than it was in prior cycles.
Below is a concise snapshot of where the largest listed miners stand today. Status reflects whether AI/HPC is already generating revenue or is still in the planning or evaluation phases, based on company disclosures and mainstream reporting.
What to watch now is simple and measurable. Track contracted AI megawatts and dollars per MW per year in new filings, utility capex trajectories, and ERCOT load revisions, and thirty-day averages for bitcoin fees relative to the subsidy using sources like CoinWarz.
Those data points will tell you how much mining power shifts to GPUs, how quickly campuses energize, and how the per-MW revenue gap evolves. The largest miners are already executing on that playbook.
