The Economics of Compute: Bitcoin Mining vs. AI Inference Costs

A recent cost breakdown shared by Ethereum co-founder Vitalik Buterin sheds light on the fundamentally different financial structures of Bitcoin mining and artificial intelligence inference workloads.

Bitcoin Mining: Electricity as the Primary Driver

The data reveals that electricity is the single largest cost component in producing a Bitcoin. The per-coin cost structure is as follows:

  • Electricity Cost: Approximately $40,600, constituting 43.3% of total cost.
  • Compute Hardware Capex: Approximately $37,600, or 40% (includes hardware, cooling, power infrastructure, and financing).
  • Labor: Approximately $3,200 (3.4%).
  • Facility Rent: Approximately $2,500 (2.7%).
  • Maintenance & Other Opex: Approximately $7,400 (7.9%).

In essence, capital expenditure on equipment and ongoing power consumption account for over 80% of mining costs combined, making electricity price volatility a critical factor for profitability.

AI Inference: Hardware Investment Dominates

The cost profile for AI inference presents a stark contrast. Calculated on a per-GPU-year basis:

  • Hardware Capex: Approximately $19,900, a dominant 76% of the total.
  • Electricity Cost: Only $1,050, a mere 4.0%.
  • Labor: Approximately $1,400 (5.3%).
  • Facility Costs: Approximately $900 (3.4%).
  • Other Expenses: Approximately $2,800 (10.7%).

For AI inference, the initial investment in high-performance computing hardware—like GPUs or specialized chips—is the overwhelming cost driver, with ongoing energy consumption playing a minor role.

Understanding the Core Disparity

As highlighted by Buterin, this divergence stems from their operational paradigms. Bitcoin mining is a continuous, compute-intensive competition requiring dedicated hardware to run 24/7, translating power into a massive, recurring operational expense.

AI inference, while also compute-heavy, often involves intermittent or on-demand processing tasks. Therefore, the performance, efficiency, and upfront cost of the hardware itself—not sustained energy draw—define its total cost of operation. This structural difference significantly impacts each industry's sensitivity to energy prices and influences global operational and siting strategies.