Ethereum co-founder Vitalik Buterin has drawn attention to a fundamental difference in the operating economics of Bitcoin mining and artificial intelligence (AI) inference: electricity makes up nearly half of the cost of mining a Bitcoin, while computing infrastructure dominates the expense of running AI models.

In an Oct. 17 post on X, Buterin compared the cost structures of the two industries, presenting estimates that show Bitcoin mining expenses split roughly evenly between hardware capital expenditure (CapEx) and power, while AI inference costs are overwhelmingly concentrated in computing equipment.

Bitcoin Mining: Costs Split Between Hardware and Power

According to Buterin, Bitcoin mining expenses are divided approximately 50/50 between computing CapEx and electricity. That balance reflects the demands of industrial-scale mining: operators must purchase specialized hardware — application-specific integrated circuits (ASICs) built solely for mining — and maintain continuous power supplies as they compete for block rewards.

The cost breakdown accompanying the post estimates total mining expenses at approximately $93,700 per Bitcoin. Computing infrastructure accounts for $37,600, or roughly 40% of the total.

Electricity consumed directly by mining machines is estimated at $40,600 per Bitcoin. Additional power-related expenses contribute another $2,400, bringing combined electricity costs to $43,000 — 45.9% of the total.

Other line items include approximately $3,200 for labor, $2,500 for facility rent, and $7,400 for maintenance, insurance, taxes, bandwidth, and related operating requirements.

The figures underscore why electricity prices remain a critical driver of mining profitability, though actual production costs vary with hardware efficiency, local energy rates, financing arrangements, and operating conditions. And because mining hardware is single-purpose and cannot be redeployed to other workloads, that cost mix ties operators' economics closely to where — and at what price — they can secure power.

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

AI Inference: Computing Infrastructure Dominates

Buterin contrasted this with AI inference — the process by which trained models generate responses, predictions, and other outputs. In that sector, computing capital expenditure accounts for approximately $19,900 per GPU-year, representing 76% of total inference costs.

Electricity plays a far smaller role. Direct power consumption is estimated at $1,050 per GPU-year, with other electricity expenses adding $210. Combined, these categories represent roughly 4.8% of the estimated $26,260 annual cost per GPU.

Labor contributes another $1,400, while facility rent accounts for $900. Remaining expenses — including maintenance, insurance, taxes, and bandwidth — total approximately $2,800.

Unlike mining machines, the GPUs used for inference are general-purpose processors that can serve a wide range of computing workloads. With equipment rather than power dominating the ledger, hardware pricing, financing, and utilization are the variables that weigh most heavily on operators.

Divergent Economic Pressures

The comparison highlights the different cost pressures facing each industry. Bitcoin miners must balance equipment investments against substantial recurring electricity bills, whereas AI operators carry greater exposure to computing infrastructure costs.

Because the two breakdowns use different units — cost per Bitcoin mined versus cost per GPU-year — the figures speak to cost composition rather than a direct, like-for-like expense total.

These dynamics help explain why some Bitcoin mining companies are expanding into AI data centers. Existing power agreements and infrastructure can support such diversification, though financing, equipment requirements, and customer demand remain important considerations. What carries over in such conversions is largely site-level infrastructure — grid connections, power contracts, and cooling — rather than the mining machines themselves.

But's comparison does not establish that either industry is universally cheaper to operate. Rather, it illustrates how hardware investment and electricity consumption weigh differently across the two operating models — a split that points to different variables for readers to watch: energy markets on the mining side, and the supply and pricing of computing hardware on the AI side.