Ethereum co-founder Vitalik Buterin compared the cost structures of Bitcoin mining and artificial intelligence inference in an Oct. 17 post on X, sharing two tables that itemize spending in each industry. According to his figures, electricity represents 45.9% of Bitcoin mining costs but only 4.8% of AI inference costs, while computing hardware and related capital spending accounts for roughly 76% of AI inference expenses.

The comparison arrives as several public mining companies — including MARA, TeraWulf, Core Scientific and Hyperscale Data — build out AI and high-performance computing capacity alongside their crypto operations. TeraWulf holds a 20-year lease with Anthropic, and Core Scientific has an AI data center deal with AMD.

Fun fact of the day: Bitcoin mining cost is split about 50/50 between compute capex and electricity. Meanwhile, AI inference is almost completely dominated by compute capex. pic.twitter.com/PCFKZSuNRw

— vitalik.eth (@VitalikButerin) October 17, 2026

Buterin said Bitcoin mining costs are split about evenly between computing investment and electricity, while AI inference is almost completely dominated by computing capital expenditure.

How Bitcoin Mining Costs Break Down

The mining table lists a total of about $93,700 per Bitcoin. Computing capital expenditure makes up about $37,600, or around 40% of that total. Electricity supplied directly to machines costs $40,600 per Bitcoin, or 43.3%, and other electricity costs add $2,400. Taken together, electricity accounts for 45.9% of the overall figure.

Labor costs $3,200 per Bitcoin, and facility rent runs about $2,500. Other expenses, such as maintenance, insurance, taxes and bandwidth, come to roughly $7,400.

The computing category is broad: it includes hardware, cooling, power equipment and financing, meaning its scope goes beyond the purchase price of mining machines.

AI Inference Costs Lean on Hardware

The AI table is denominated in dollars per GPU-year, with total costs of about $26,260. Computing capital expenditure makes up roughly $19,900, or 76% of that total, while electricity accounts for 4.8% when all categories are combined.

Labor adds $1,400 per GPU-year, facility rent about $900, and other operating costs roughly $2,800.

Because one table is measured per Bitcoin and the other per GPU-year, the dollar totals cannot be compared directly. The figures cover costs rather than revenue, so they describe how each type of operation spends, not what it earns. The tables also do not set a uniform cost level for every miner. Operators with flexible power contracts can scale back when hashprice — a widely used gauge of daily mining revenue per unit of hash power — falls, while AI operators have less room to adjust, because financed hardware keeps costing money whether or not it is fully used.

Miners Expand Into AI

Several miners have moved toward AI through recent deals. Core Scientific signed an AI data center agreement with AMD.

TeraWulf signed a 20-year lease with Anthropic at its Hawesville, Kentucky campus. The site is expected to support about 401 megawatts and an estimated $19 billion in revenue over the initial term, with initial capacity expected in the second half of 2027. Under the arrangement, TeraWulf supplies power and facilities, while customers handle servers and processors.

Hyperscale Data sold about 100 Bitcoin and set up a Bitcoin-backed credit facility to fund construction in Michigan. Its first deployment covers about 20 megawatts under a ten-year agreement.

MARA agreed in July to buy a 1,200-acre Texas property for high-performance computing and Bitcoin mining. The site is expected to offer one gigawatt of grid capacity by October 2027 and two gigawatts by April 2028.

Across the group, the AI arrangements build on what the miners already hold — power access, sites and electrical capacity — with TeraWulf's power-and-facilities split the most explicit example of the structure. The nearest scheduled checkpoints for these buildouts fall in 2027 and 2028, starting with TeraWulf's initial Anthropic capacity and MARA's first Texas grid capacity target.