Vitalik Buterin: Bitcoin Mining Costs are Roughly 50/50, AI Inference Relies Almost Entirely on Equipment Capital
Ethereum co-founder Vitalik Buterin stated that approximately half of the costs of Bitcoin mining come from capital expenditures on computing equipment, while the other half comes from electricity consumption. In contrast, the costs of AI inference are almost entirely dominated by capital expenditures on computing power.
His breakdown of costs shows that capital expenditures for mining equipment (including hardware, cooling, power supply, and financing) and electricity costs are roughly equal; in some more detailed estimates, electricity accounts for about 43% of the costs associated with each Bitcoin, while equipment capital accounts for about 40%. For AI inference, capital expenditures can account for about 76% on a per-GPU annual basis, with electricity costs only about 4%. This structural difference makes mining profitability highly dependent on electricity prices, mining machine efficiency, Bitcoin prices, and network difficulty, while AI inference is more reliant on GPU procurement costs, depreciation, and utilization rates.
As demand for AI computing power grows, some Bitcoin mining companies are shifting their existing power capacity and data center facilities towards AI and high-performance computing. Reports indicate that companies like IREN, TeraWulf, Core Scientific, and Riot Platforms have signed large-scale AI/HPC hosting or computing power contracts, with a total contract value exceeding $100 billion, but the actual capacity that has begun billing is only about 550 megawatts, far below the over 4 gigawatts contracted.
Transformation requires additional capital. Industry estimates show that completing the planned AI infrastructure requires nearly $50 billion in near-term funding gaps; some companies, such as IREN, face a gap of about $21.1 billion, while already signing a $9.7 billion contract with Microsoft and guiding for high capital expenditures. Some companies are reducing computing power in the first half of 2026, recording asset impairments, and reallocating electricity.
The mining business is directly exposed to fluctuations in electricity prices and Bitcoin prices, while the AI computing business bears the risks of GPU procurement, depreciation, and underutilization. Both types of businesses compete for the same electricity and computing resources, and ultimately profitability depends on how resources are converted into revenue; currently, the backlog of AI contracts has become a narrative for mining company valuations, but the pace of delivery and funding availability remain key variables.
Source: Public Information
ABAB AI Insight
Vitalik Buterin has long focused on energy and computing economics, having analyzed the marginal costs and security of PoW multiple times, and has continued to discuss the resource competition between mining and other computing uses since Ethereum's transition to PoS; his cost breakdown this time continues the distinction between "electricity as a continuous marginal cost" and "hardware as sunk capital."
In terms of capital pathways, mining companies are treating low-cost electricity access and data center space as transferable assets, signing multi-year hosting contracts with Microsoft, CoreWeave, Anthropic, etc., reallocating power capacity originally intended for ASICs to GPU clusters; the motivation is that mining profits will be compressed after the 2024 halving, while AI demand pays higher unit electricity rents, forming a cash flow transition from "price-driven" to "contract-driven."
Similar to the path of cloud providers in the 2010s from building their own data centers to hosting excess capacity, current mining companies are in a phase of expansion from "dedicated mining" to "hybrid computing infrastructure"; unlike early mining operations that relied solely on electricity price arbitrage, they now need to manage GPU supply chains, cooling upgrades, and tenant credit risks simultaneously.
This essentially represents a restructuring of the industry chain: electricity and space are shifting from Bitcoin-specific assets to general computing infrastructure, with pricing power moving from simple electricity prices and difficulty to GPU utilization rates and long-term contracts; the mechanism at play is that the halving has reduced the ongoing cash flow attractiveness of mining, while AI capital expenditures have pushed marginal costs almost below electricity prices, making existing electricity access a scarce resource, prompting mining companies to complete asset repricing through impairments and reallocations.
ABAB News · Cognitive Law
Electricity is a continuous tax, equipment is a sunk bet
Utilization determines life and death, electricity price only sets the threshold
Contracts are leverage, delivery is revenue