Top Bitcoin Miners Turn to AI Revenues as Bear Market Squeezes Profit Margins

• 70% of Bitcoin miners turn to AI computing services to boost their revenue amid a prolonged bear market.

• Bitcoin and the broader crypto market have experienced a massive drop since September.

AI Revenue Becomes Critical for Bitcoin Miners

Nearly 70 percent of the world’s largest Bitcoin miners are now generating income from artificial intelligence or high-performance computing services. This reflects a critical strategic shift as companies look to withstand prolonged price uncertainty, network difficulty growth, and thinning margins.

Where miners were once valued primarily by hashrate expansion, today’s competitive edge increasingly lies in megawatts that can be converted into contracted GPU-based revenue. Power access, substation capacity, and data-center readiness are now defining scale as much as ASIC deployments.

TeraWulf Establishes Benchmark for Hosting Economics

TeraWulf has emerged as a reference point for the hybrid model after signing two decade-long hosting deals with Fluidstack totaling about 200 megawatts at its Lake Mariner site. According to Barron’s, Google is supporting a portion of Fluidstack’s lease obligations, up to roughly $1.8 billion, in exchange for warrants equal to approximately 8% of TeraWulf.

The implied economics suggest annual revenue of about $1.85 million per megawatt of hosting capacity. This figure has become the new benchmark for miners positioning themselves to attract major AI customers. These economics compare favorably with returns from running modern ASICs at full power, encouraging others to pursue similar models.

Major Mining Firms Expand AI and HPC Infrastructure

Core Scientific has expanded its 12-year partnership with CoreWeave by adding around 70 megawatts of HPC capacity expected to be operational in the second half of 2025. Bitdeer continues to operate commercial AI cloud services using NVIDIA’s DGX hardware, while Iris Energy has reported revenue from cloud workloads powered by H100 and H200 GPUs.

Beyond active deployments, companies are preparing infrastructure to scale further. CleanSpark acquired more than 270 acres and secured about 285 megawatts of long-term power capacity in Texas for what it describes as a next-generation AI and HPC campus. Marathon agreed to acquire a 64 percent stake in Exaion, an EDF subsidiary, to expand global AI and HPC capabilities, with an option to increase its stake to 75 percent by 2027.

Mining Expansion Slows as Capacity Shifts to AI

Other firms are actively reshaping previously exclusive bitcoin-mining sites. Riot is evaluating converting roughly 600 megawatts at its Corsicana campus to support AI or HPC, leading the company to revise its year-end 2025 hashrate forecast downward from 46.7 exahash to 38.4 exahash. Bitfarms has begun marketing its properties to AI customers and initiated a feasibility study on its future roadmap.

Cipher Mining is reported to have a multi-year hosting arrangement with Fluidstack supported by Google-linked commitments, though details are not consolidated into a single disclosure. Abu Dhabi-based Phoenix Group is preparing to scale data-center capacity beyond 1 gigawatt, with a heavy AI footprint, and is exploring a U.S. listing to support capital needs.

AI Economics Outperform ASIC Economics

The economic drivers behind this pivot are clear. Under current network conditions, about 1.08 to 1.10 zettahash per second. Equally, a daily block production of 144 and average fees ranging from 0.3 to 2 BTC per block, with one megawatt of modern ASIC miners operating at around 17 J/TH, translates to roughly 0.059 exahash.

That share of the Bitcoin network produces approximately $1.0 to $1.6 million in annual gross revenue per megawatt at a Bitcoin price near $104,000, based on CoinWarz data. Even the midpoint of roughly $1.2 to $1.3 million trails the $1.85 million per megawatt implied by TeraWulf’s hosting contract.

Although margins still depend heavily on power pricing, capex, and utilization, AI hosting benefits from contractual visibility, insulating operators from fee volatility and subsidy dynamics.

Power Demand Reshapes the U.S. Data-Center Landscape

A powerful macro trend is reinforcing this reallocation. McKinsey research suggests U.S. data-center electricity demand could reach approximately 606 terawatt-hours by 2030 as AI usage scales. In Texas, ERCOT projects record demand growth over the next five years, with data-center peak load potential rising to around 35 gigawatts by 2035.

Utilities are adjusting capital planning to meet these needs. Reuters reported that American Electric Power has boosted its five-year capital plan to around $72 billion as it manages a pipeline of customer-backed contracts and more than 190 gigawatts of requests. These dynamics support miners’ claims that their existing land, substations, and interconnection rights now serve as prime entry points into the data-center market, valued for more than just exahash production.

Valuation Metrics Shift Toward Contracted Revenue

As miners redirect power toward GPUs, headline hashrate growth is no longer a singular marker of progress. In many cases, enterprise value improves when capacity shifts to AI because contracted revenue provides dependable cash flow with longer-dated commitments.

Core Scientific’s CoreWeave buildout, CleanSpark’s Texas campus, and Marathon’s Exaion expansion all reflect this multipurpose campus model, where ASIC mining, GPU hosting, and traditional colocation can share infrastructure. In this environment, the competitive differentiator becomes the ability to secure power, cooling, water, permitting, and network connectivity rather than merely scaling ASIC fleets.

Bottlenecks and Fee Shifts May Influence Future Allocations

Despite momentum, the transition is constrained by long interconnection timelines, transformer shortages, and delays in the construction of gas turbines for peaker plants. GPU supply remains unpredictable as NVIDIA ramps its Blackwell generation and hyperscalers reserve priority volumes.

On the cryptocurrency side, a potential sustained rise in average transaction fees could narrow the earnings gap. Gains of about 0.5 BTC per block would translate into roughly $200,000 to $300,000 in additional annual mining revenue per megawatt under current pricing and share-of-network assumptions. While meaningful, that still leaves AI hosting in a favorable position given the predictability of contracted income.

AI Megawatts Become the New Leaderboard Metric

Investors have begun tuning their analysis to track contracted AI megawatts and revenue per megawatt rather than prioritizing exahash. The emerging reference range for U.S. high-density hosting appears to sit near $1.5 to $2.0 million per megawatt per year, anchored by TeraWulf’s disclosed figures.

As a result, the traditional mining leaderboard is being rewritten. Companies integrating AI workloads are no longer viewed as diversifying on the margins. Still, they are increasingly seen as defining the sector’s next operating model at the crossroads of power infrastructure, compute, and digital assets.