Ethereum co-founder Vitalik Buterin has compared the costs of Bitcoin mining with AI inference, highlighting how much each activity spends on electricity and computing equipment.
In an Oct. 17 post on X, Buterin shared estimates showing that electricity accounts for 45.9% of Bitcoin mining costs, while computing capital expenditure makes up about 76% of AI inference costs.
Both estimates include hardware, cooling systems, power equipment and financing under computing capital expenditure. The figures suggest that buying and maintaining computing infrastructure can be a major expense for both industries, although the cost breakdown differs significantly.
Electricity accounts for nearly half of Bitcoin mining costs
Buterin described Bitcoin mining expenses as being split roughly evenly between computing investment and electricity.
The figures accompanying his post estimate that electricity used directly by mining computers costs $40,600 per BTC, representing 43.3% of the total. Other electricity expenses add another $2,400, or 2.6%.
Together, electricity-related costs reach $43,000 per BTC, accounting for 45.9% of the estimated total.
Computing capital expenditure adds approximately $37,600 per BTC, or around 40%. This category includes mining hardware, cooling, power equipment and financing costs rather than just the purchase price of the machines.
Other expenses include $3,200 per BTC for labor, about $2,500 for facility rent and approximately $7,400 for maintenance, spare parts, insurance, taxes, bandwidth and other operating costs.
The listed amounts add up to an estimated $93,700 per BTC. However, this is the total shown in the cost breakdown Buterin shared, not a standard production cost for every mining operation. Actual expenses depend on electricity prices, hardware efficiency, financing and facility conditions.
AI inference costs are dominated by computing equipment
The AI inference estimate shows a different cost structure. Computing capital expenditure accounts for approximately $19,900 per GPU-year, or 76% of the total.
As with Bitcoin mining, this category includes hardware, cooling, power equipment and financing.
Electricity used directly by computers costs an estimated $1,050 per GPU-year, representing 4% of total expenses. Other electricity costs add another $210, or 0.8%. Together, electricity accounts for approximately 4.8% of the estimated total.
Labor contributes $1,400 per GPU-year, while facility rent accounts for around $900. Maintenance, facility operations, insurance, taxes, bandwidth and other expenses add approximately $2,800.
The listed costs total about $26,260 per GPU-year.
Buterin said AI inference is almost entirely dominated by computing capital expenditure. The table supports that observation by assigning roughly three-quarters of the estimated cost to computing infrastructure.
The two tables use different units: dollars per Bitcoin for mining and dollars per GPU-year for AI inference. Their totals therefore cannot be directly compared to determine which activity is more expensive for an equivalent amount of computing work.
TeraWulf uses long-term leases to support AI infrastructure
Several U.S. companies are developing AI data centers while using different financing and operating arrangements.
TeraWulf, a company with Bitcoin mining operations, has been expanding into AI infrastructure. In July, reports said it was considering a potential $3.5 billion financing package for an AI campus in Kentucky, using possible leveraged loans and high-yield bonds.
The plans followed the company’s announced 20-year lease with Anthropic at its Justified Data campus in Hawesville. TeraWulf said the facility would support approximately 401 megawatts of critical computing capacity and estimated $19 billion in revenue over the initial contract term.
Initial capacity was expected in the second half of 2027, with full deployment targeted for early 2028. The $19 billion figure represents projected revenue over the contract’s initial 20-year period, not an upfront payment.
TeraWulf Chief Financial Officer Patrick Fleury described a model in which customers remain responsible for servers, processors and technology upgrades, while TeraWulf supplies the power and physical infrastructure.
The company has also used debt to support its broader operations. In October 2025, a subsidiary priced $3.2 billion in senior secured notes carrying a 7.75% annual interest rate and maturing in 2030.
Crypto companies are using digital assets to fund AI projects
Other companies have turned to cryptocurrency sales and borrowing to support their AI expansion plans.
In July, Quantum Solutions’ subsidiary GPT Pals Studio sold 1,000 ETH for approximately $1.903 million after transaction fees. The company said the proceeds would support its Japanese AI data center expansion plans.
U.S.-listed Hyperscale Data also monetized approximately 100 BTC and established a Bitcoin-backed credit facility to support construction in Michigan.
Its initial customer deployment covered about 20 megawatts under a 10-year agreement, with two optional five-year extensions. The company estimated more than $1.2 billion in revenue over the maximum contract term, assuming the extensions are exercised and the agreement proceeds as planned.
MARA has taken a different approach by securing land and power capacity for future high-performance computing projects in Texas.
The company announced a definitive agreement to acquire a 1,200-acre property intended for AI computing, flexible computing services and Bitcoin mining. MARA projected access to an initial gigawatt of grid capacity by October 2027, with total available capacity potentially reaching two gigawatts by April 2028.
The development was planned with Starwood Digital Ventures, with construction expected to proceed in phases subject to regulatory approvals. The seller, HIF, would retain a minority ownership interest once a high-performance computing lease was executed.
These projects show how companies that began in Bitcoin mining are exploring AI infrastructure as another source of business. Some are relying on long-term customer leases and debt, while others are selling digital assets or securing power capacity before construction.
Buterin’s comparison highlights one reason the two industries face different operating pressures: electricity represents a much larger share of the estimated cost of Bitcoin mining than it does in the AI inference breakdown. For AI infrastructure, the initial investment in computing equipment and related systems accounts for the largest share of costs.
The estimates provide a useful snapshot of the expenses involved, but actual costs vary by hardware, utilization, financing terms, electricity rates and the operating model used by each company.
