Vitalik Buterin posted a chart breaking down Bitcoin mining costs. The split looks nothing like AI. The Ethereum co-founder shared a side-by-side cost comparison in a post on X showing where money actually goes in each industry, down to line items like cooling, facility rent, and insurance. The numbers suggest the two industries, despite both depending on massive compute infrastructure, run on fundamentally different cost structures.
In an X post, Buterin wrote, “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.”
What Buterin Posted
The chart Buterin shared breaks Bitcoin mining costs down per BTC produced. Compute capex—hardware, cooling, and power gear, including financing—accounts for about $37,600 per BTC, or roughly 40% of total cost.
Electricity into the computers themselves adds another $40,600, or 43.3%. The remainder splits across smaller categories: other electricity (2.6%), labor (3.4%), facilities rent (2.7%), and other costs, including maintenance, insurance, tax, and bandwidth (7.9%).
The AI inference side of the chart tells a different story. Compute capex there runs about $19,900 per GPU-year, making up roughly 76% of total cost—nearly double Bitcoin mining’s share. Electricity into the computers accounts for just $1,050, or 4% of the total. The remaining categories—other electricity, labor, facilities rent, and other costs—are each a smaller share than their Bitcoin mining counterparts.
Why the Split Looks So Different
The gap comes down to how each system actually runs. Bitcoin mining machines operate at a near-constant, predictable power draw around the clock, which is why electricity shows up as such a large, steady share of total cost—it accrues continuously regardless of demand. AI inference workloads are far more variable: GPU utilization rises and falls with usage, but the hardware itself remains scarce and expensive, meaning the upfront capital cost of acquiring and financing the chips dominates the economics even when the GPUs sit partially idle.
Part of a Running Theme
This is not the first time Buterin has used his own numbers to push back on a Bitcoin-AI narrative. In September, he publicly rejected a prediction from AI-risk commentator Liron Shapira that artificial intelligence could crash Bitcoin’s price by more than 50% within two years by breaking its underlying cryptography.
Buterin called the odds of AI defeating Bitcoin’s hash functions or proof-of-work mechanism “tiny” and noted that roughly 90% of his net worth was effectively betting against that outcome. His cost-breakdown post reads as a quieter follow-up to the same underlying position: that Bitcoin’s infrastructure, economically and cryptographically, is built differently from the AI systems some expect to threaten it.
A Cost Structure Behind the Industry’s AI Pivot
The numbers Buterin highlighted help explain a trend already playing out across the Bitcoin mining industry: public miners have increasingly converted mining capacity toward AI and high-performance computing workloads, pointing to AI’s higher revenue potential per unit of power.
CoinShares’ Bitcoin Mining Report for the fourth quarter of 2025 found that building and operating a Bitcoin mining facility generally costs $700,000 to $1 million per megawatt, versus roughly $20 million per megawatt for comparable AI infrastructure. That spread is already reshaping the public mining sector. Core Scientific’s pivot into AI colocation helped drive a $9 billion acquisition deal with CoreWeave, while Riot Platforms has earmarked 600 megawatts of its Corsicana, Texas site for AI and high-performance computing, pausing further Bitcoin mining buildout there in the process.
Also Read: Vitalik Wraps AI Agents in Tor + zkAPI for True Paranoid Mode
