In this article4 sections
Electricity accounts for 45.9% of Bitcoin mining expenses in a cost table Ethereum co-founder Vitalik Buterin shared, while computing capital expenditure takes roughly 76% of AI inference costs, according to Crypto.news. Buterin published the comparison in an Oct. 17 post on X.
The Ethereum creator framed the gap as a matter of where the money goes, not how much each industry spends. In the post he wrote that mining costs divide about evenly between hardware and power, whereas inference is “almost completely dominated by compute capex,” a phrase Crypto.news quoted. Cryptobriefing reported that the exercise amounts to comparing how each industry structures its bills.
Also readVitalik Buterin Details 'Paranoid Mode' AI Agent on Ethereum
Key facts
- The mining table lists $40,600 per BTC for electricity to computers (43.3%) plus $2,400 (2.6%) in other power costs, for a combined 45.9%.
- Compute capex in mining is about $37,600 per BTC, around 40%, covering hardware, cooling, power equipment and financing.
- In AI inference, compute capex is roughly $19,900 per GPU-year, or 76% of an estimated $26,260 total.
- Buterin posted the comparisons on X on Oct. 17, 2026.
- Cryptobriefing reported facility-level estimates of $700,000 to $1 million per megawatt to build and run a Bitcoin mining site, against about $20 million per megawatt for AI infrastructure.
Two tables, two very different splits
The mining breakdown, denominated in dollars per Bitcoin, adds labor at $3,200, facility rent near $2,500 and about $7,400 of other expenses — maintenance, spare parts, operations, insurance, taxes and bandwidth — for roughly $93,700 in total. Crypto.news noted the figure describes the table itself and does not set a uniform production cost for miners.
The AI table uses a different unit, dollars per GPU-year. Power to computers is $1,050 (4%) and other electricity $210 (0.8%), for 4.8% combined. Labor adds $1,400 and rent about $900, with roughly $2,800 in remaining maintenance, operating, insurance, tax and bandwidth costs. Because the units differ, Crypto.news observed, the dollar totals cannot be read as the cost of doing the same task.
Also readButerin: Hegota will be Ethereum's last 'normal' fork
The outlets describe the electricity share of inference slightly differently: Crypto.news reports 4.8% combined, while Cryptobriefing gives a range of 4% to 4.8% and puts the dollar figure somewhere between $1,050 and $1,260.
Why it matters
For miners, a power bill that makes up nearly half of estimated costs means cheap, stable electricity remains a structural advantage, and exposure to volatile power markets is a direct margin risk. AI operators face the opposite pressure: with roughly 76% of inference costs tied to chips and facilities, keeping expensive GPUs busy carries more weight than the power bill.
The comparison lands as both industries compete for the same grid connections, cheap power and physical space. Buterin’s own network, Ethereum, moved off proof-of-work years ago, so he is commenting on energy economics his chain no longer pays for.
What to watch
The facility-level gap — $700,000 to $1 million per megawatt for mining against about $20 million for AI, as Cryptobriefing reported — is the number to track as miners weigh hosting or converting sites. Two arrangements to follow are TeraWulf’s 20-year Anthropic lease at its Justified Data campus in Hawesville, Kentucky, which was expected to begin capacity in the second half of 2027, and MARA’s planned 1,200-acre Texas site, where an initial gigawatt of grid capacity was projected by October 2027.
This article is not financial advice; cryptocurrency and AI infrastructure markets are volatile and uncertain.
Sources: crypto.news, Cryptobriefing
