Seven of the top 10 hashrate companies are already generating revenue from AI or high-performance computing efforts, with three others reporting plans to follow suit.
This shift combines miners’ energetic land and interconnects with contract revenue from GPU customers, creating a second line of business that competes with ASICs running at full power.
AI partnerships redefine mining economics and investor focus
TeraWulf has established a benchmark totaling approximately 200 MW at Lake Mariner after signing two 10-year hosting agreements with Fluidstack.
According to Barron’s, Google is backing a portion of Fluidstack’s lease obligations (up to about $1.8 billion) and has received warrants worth about 8% of TeraWulf. Disclosed transaction calculations suggest headline revenue for the current period will be approximately $1.85 million per MW per year, which many miners use as a benchmark when courting AI tenants.
Core Scientific is expanding its 12-year relationship with CoreWeave for approximately 70 MW of additional HPC capacity, targeted for commissioning in the second half of 2025. Bitdeer continues to operate a commercial AI cloud based on NVIDIA DGX systems, and Iris Energy reports an AI cloud business running on H100 and H200 GPUs.
Some companies are building real estate for the next wave. CleanSpark announced on October 29 that it has secured 271 acres of land and approximately 285 MW of long-term power in Texas for what it calls its next-generation AI and HPC campus. In August, Marathon agreed to acquire a 64 percent stake in EDF subsidiary Exaion, with an option to increase its stake to 75 percent by 2027, to expand its global AI and HPC capabilities.
Riot is evaluating the conversion of approximately 600 MW for AI or HPC in Corsicana and has paused some of its mining expansions, resulting in a lower hashrate guidance for the end of 2025 from 46.7 EH/s to 38.4 EH/s. Bitfarms has hired consultants to conduct feasibility studies and market its site to AI clients.
Cipher Mining reportedly has a multi-year Fluidstack agreement with Google with a linked lease agreement, but not all terms were disclosed in one primary filing. Abu Dhabi’s Phoenix Group has hinted at plans to scale up alt=””>
Economic problems are about power and predictability.
With the current network context of about 1.08-1.10 ZH/s, 144 blocks per day, and fees ranging from about 0.3-2.0 BTC per block, 1 MW of modern ASICs of about 17 J/TH corresponds to a hashrate of about 0.059 EH/s.
According to CoinWarz price and hashrate data, the network’s shares earn about $1 million to $1.6 million per MW per year in total mining revenue, excluding power and operating costs, which equates to a Bitcoin price of about $104,000. The midpoint of that range, about $1.2 million to $1.3 million, is below the $1.85 million per MW per year implied by TeraWulf’s AI contract.
In both models, margins are determined by power prices, capital expenditures (capex), and utilization rates. Still, the contractual nature of AI hosting has become an important feature for equity investors seeking a more stable cash flow rather than pure risk and fee exposure.
Analysis shows peak demand of about 35 GW, with ERCOT projects recording peak demand over the next five years, with data centers being the main driver, according to McKinsey’s graph. These numbers are consistent with miners’ claims that the power grid, substations, and land banks are insufficient to supply not only Exahash but also the AI campus.
This rebalance changes what is important within the mining league table.
Miners that direct new megawatts toward AI are likely to record lower headline hashrate increases than pure operations. However, its value can increase through contract revenues, power options, and long-term contracts.
Core Scientific’s addition of CoreWeave gives this model a 12-year stamp. CleanSpark’s 285 MW plan and Marathon’s acquisition of Exaion will encourage miners to own and operate mixed-use campuses where GPUs, miners, and in some cases standard colocation can share infrastructure. Riot’s public evaluation of 600 MW in Corsicana shows how quickly configurations can change when a site already has transformers, switchgear, water rights, and fiber optic infrastructure in place.
There are restrictions. ERCOT’s interconnection schedule, gas turbine availability for new peakers, and transformer lead times all determine how quickly high-density halls can be powered. GPU supply will continue to be a variable as Blackwell and successor parts increase and hyperscalers allocate inventory to in-house builds.
On the cryptocurrency side, changes to the pricing regime that significantly increase fees per block could close some of the revenue per MW gap between mining and AI hosting. Based on the simple network occupancy calculation above, moving around 0.5 BTC per block with a sustained average fee is worth around $0.2-0.3 million per MW per year in total revenue for miners at current price levels.
Investors are paying attention not only to Exahash but also to the revenue structure.
Contracted AI megawatts and annual dollar amounts per MW are becoming new disclosures to track. A range of $1.5 million to $2 million per MW per year has emerged as a practical benchmark for high-density hosting in the US, with the numbers disclosed by TeraWulf serving as the current standard.
Updates to utility capital spending plans and interconnection queues are now as relevant to miner prospects as ASIC delivery schedules. As U.S. spot power tightens, miners with already energized land, permitted pads, and backup substations can monetize their options faster than greenfield entrants.
The international angle adds weight. The move by Marathon and Exion will connect U.S. miners with EDF affiliates in the French power system and link GPU hosting to energy assets adjacent to the state.
Phoenix Group’s plans to scale up in the Gulf region include incorporating sovereign power economics into its AI infrastructure as it considers a U.S. listing.
Such structures could draw more miners into joint ventures where utilities and energy investors enter into long-term contracts in exchange for capacity rights, preferential interconnections and equity stakes.
In terms of cryptocurrency fundamentals, the pivot could slow the rate of network hashrate expansion until 2026 if a significant portion of new power is routed to GPUs rather than ASICs. As new sites come online and older fleets are updated, the network will continue to add hashes, but the slope may flatten compared to the last spike.
The entry of capital into mining will not be stopped, as profits can still improve due to rising Bitcoin prices and rising fees. However, the hashrate leaderboard is a weaker proxy for stock value than in previous cycles.
Below is a concise snapshot of the current status of the largest listed miners. Status reflects whether AI/HPC is already generating revenue or is still in the planning or evaluation stage, based on company disclosures and mainstream reporting.
What to watch now is simple and measurable. Use sources like CoinWarz to track annual contracted AI megawatts and dollars per MW in new applications, utility capital spending trajectory, ERCOT load revisions, and 30-day averages of Bitcoin fees for subsidies.
These data points tell us how much mining power moves to GPUs, how quickly the campus powers up, and how the revenue gap per MW changes. The biggest miners are already implementing that playbook.
