Ethereum co-founder Vitalik Buterin compared Bitcoin mining with AI and shared a cost estimate: electricity accounts for 45.9%, while computational capital expenditure accounts for about 76% of the total reasoning costs.

• In the comparison shared by Buterin, electricity costs account for 45.9% of the total mining costs for Bitcoin.
• Calculated capital expenditure accounts for approximately 76% of the estimated cost of AI.
• Both tables include refrigeration, power supply equipment, and financing in the calculation of capital expenditures.
• US operators advance the construction of AI facilities through customer leases, loans, and the sale of crypto assets.

In a post on X on October 17th, Buterin indicated that the cost of Bitcoin mining is roughly split equally between computational investment and electricity, while AI inferred that it is almost entirely dominated by computational capital expenditure.

The electricity cost for Bitcoin production accounts for nearly half of the total cost.

The post with Buterin distinguishes between the expenses required for purchasing computing infrastructure and the ongoing costs of powering it.

He wrote, "The cost of Bitcoin mining is approximately 50% computing capital expenditure and 50% electricity."

The mining spreadsheet attached to his post provides a more detailed breakdown. The electricity cost directly for powering the computers is estimated at $40,600 per BTC, which accounts for 43.3% of the total cost. Additional electricity expenses amount to $2,400, accounting for 2.6%.

The total cost of these two types of electricity amounts to $43,000 per BTC, which constitutes 45.9% of the displayed costs.

The calculated capital expenditure is approximately $37,600 per BTC, which accounts for about 40%. This category encompasses hardware, cooling, and power supply equipment, including financing, so the scope is not limited to just the purchase price of mining machines.

The remaining categories include an labor cost of $3,200 per BTC, a venue rental fee of about $2,500, and other expenses amounting to approximately $7,400. Maintenance, spare parts, facility operations, insurance, taxes, and bandwidth are all included in the last category.

After adding up the listed amounts, the estimated total cost is approximately $93,700 per BTC. This calculation refers to the table shared by Buterin and does not imply that the production costs for all Bitcoin miners are the same.

AI Inference expenditures are concentrated on computing infrastructure

The table uses different units, namely per GPU dollars per year, and allocates a larger proportion of expenses to calculating capital expenditures.

This estimate sets this category at approximately $19,900 per GPU, which accounts for 76% of the total cost. Similar to the mining table, this category also includes hardware, cooling, power supply equipment, and financing.

The electricity directly supplied to the computer costs $1,050 per GPU year, which accounts for 4%. Other electricity expenses amount to $210, accounting for 0.8%, for a total electricity cost of 4.8%.

The labor cost increases by $1,400 per GPU year, and the venue rental is about $900. The total expenses for maintenance, operations, insurance, taxes, and bandwidth amount to approximately $2,800.

The total amount listed is approximately $26,260 per GPU year. Since one table is billed per Bitcoin and the other table is billed per GPU year, the total dollar amounts of the two cannot be directly compared to determine the cost of performing the same task.

Buterin wrote when summarizing this difference, and AI inferred that it is 'almost entirely dominated by capital expenditure on computing.' A detailed breakdown shows that about three-quarters of the total inference cost falls into this category, while electricity, labor, rent, and other operating expenses make up the remainder.

TeraWulf separates the custody of AI from the ownership of the processor.

The American data center protocol demonstrates how companies allocate responsibilities for facilities and computing equipment.

According to previous reports from crypto.news, TeraWulf considered a potential $3.5 billion financing in July to support its AI campus located in Kentucky. Bloomberg stated that these financing plans included possible leveraged loans and high-yield bonds.

Prior to the financing plan, TeraWulf announced a 20-year lease agreement with Anthropic for a location in their Justified Data campus in Horswell. The company stated that this facility will support approximately 401 megawatts of critical computing power and is expected to generate $19 billion in revenue during the initial contract period.

The initial production capacity is expected to be launched in the second half of 2027, with a full deployment target set for early 2028. The revenue estimates cover the 20-year period of this agreement and do not represent prepaid payments.

Chief Financial Officer Patrick Fleury stated that the operating model is such that customers continue to be responsible for server, processor, and technology upgrades, while TeraWulf provides power and physical infrastructure.

Previous financings have supported its New York operations. In October 2025, a subsidiary of TeraWulf issued $3.2 billion in senior secured notes with a coupon rate of 7.75%, maturing in 2030.

Encrypted asset sales and power supply agreements support the development of AI.

Other companies are also utilizing digital assets to finance their AI data center projects. A report on July 31 documented the transactions announced the previous day by Quantum Solutions and Hyperscale Data, which is listed in the United States.

Quantum's subsidiary, GPT Pals Studio, sold 1,000 ETH. After deducting transaction fees, they received $1.903 million, which was used to support the expansion plan of Japan's AI data center. Hyperscale realized approximately 100 BTC and established a credit arrangement secured by Bitcoin for construction projects in Michigan.

The initial customer deployment of Hyperscale covers approximately 20 megawatts, with a duration of 10 years, and there is an option to extend it for two additional 5-year periods. The company estimates that revenue will exceed 1.2 billion US dollars over the longest term, provided that these extension options are exercised.

The focus of the announcement in July is to secure land and power capacity for future development in Texas. The company disclosed a final agreement to acquire a 1,200-acre property, which is planned to be used to support high-performance computing, flexible computing services, and Bitcoin mining.

According to the announcement from MARA, it is expected that this plot of land will be able to connect to an initial grid capacity of 1 gigawatt by October 2027, and by April 2028, the total available capacity is expected to reach 2 gigawatts.

MARA plans to jointly develop this property with Starwood Digital Ventures. Phase-by-phase construction requires regulatory approval. The seller, HIF, will retain a minority stake after signing the HPC lease agreement.