Bitcoin Mining Estimate Puts Electricity at 45.9% of Total Costs

Bitcoin (BTC) mining’s estimated cost mix puts electricity at nearly half the total, while an AI inference estimate assigns most costs to computing equipment, as Core Scientific converts a Texas campus for AI use.

Ethereum co-founder Vitalik Buterin shared the estimates Oct. 17. His comparison puts the cost of mining one BTC at about $93,700, including $43,000 for electricity and $37,600 for computing capital expenditure, or compute capex. Electricity represents 45.9% of the mining estimate, while compute capex accounts for about 40%.

The electricity line includes $40,600 for power used by mining machines and $2,400 in other electricity costs. The estimate also assigns $3,200 to labor, $2,500 to facility rent and $7,400 to other costs. Together, those categories bring the estimate to $93,700 per BTC mined.

The AI inference estimate totals about $26,260 per GPU-year. It assigns $19,900, or 76%, to compute capex and $1,260, or 4.8%, to electricity. That electricity total consists of $1,050 for power used by computers and $210 in other electricity costs.

The remaining AI estimate assigns $1,400 per GPU-year to labor, $900 to facility rent and $2,800 to other costs. A GPU-year measures costs for one GPU over a year of inference; the mining figure uses one BTC produced as its output. The totals therefore do not show that one activity costs more overall than the other.

The estimates divide each activity’s costs into categories, but they are not audited costs for a representative Bitcoin miner or AI operator. In the AI estimate, computing capital accounts for a much larger share than electricity.

Core Scientific has begun converting its Pecos, Texas, campus into an AI data center. The site had 300 megawatts of gross power capacity in use for Bitcoin mining. Initial capacity is expected in early 2027.

The conversion is part of a broader shift in mining companies’ use of data center capacity for AI workloads. Core Scientific’s project provides a concrete example of a mining campus being repurposed, while Buterin’s estimates show a different expense mix for AI inference.