Ethereum Co-Founder Vitalik Buterin Says Bitcoin (BTC) Mining Costs Split Nearly 50-50 With Power

Hardware and Power Split Mining Costs Nearly Evenly

Vitalik Buterin, positioned as Ethereum co-founder and a direct participant in the economics of blockchain networks, says the cost structures of Bitcoin (BTC) mining and artificial intelligence inference are fundamentally different. In a post on X on 17 October, he presented the difference between the two cost models as an “interesting fact,” and the comparison has since drawn argument across mining and AI infrastructure circles. Both technologies depend on computing resources, he noted, yet the share of the bill carried by hardware and by electricity is not alike in the two industries. His breakdown puts the cost of crypto mining almost evenly on hardware investment and electricity. Hardware outlays are capital expenditure: money spent up front on specialized machines built for the network's Proof of Work design. Electricity, by contrast, is a recurring operating cost, and it arrives continuously for as long as the machines stay online. Miners, in his description, commit large sums to dedicated equipment while simultaneously paying substantial power bills to run it. The comments reached a market already attentive to the Bitcoin price and to the cost base underneath miner margins, and the framing quickly spread through trading and infrastructure communities on the platform. Inference, the stage at which a trained AI model generates answers or completes tasks, sits on the other side of his ledger. His claim is that hardware investment absorbs nearly all of inference spending, because electricity consumption takes a comparatively small share. The distinction matters because the two industries are often discussed as if they compete on identical terms for the same resources. The comparison separates them at the level of the invoice: one business pays roughly half its costs to the power grid, the other pays almost nothing there and nearly everything to the equipment maker.

The arithmetic behind the post breaks into two halves. On the Bitcoin mining side, hardware investment and electricity bills each take close to half of total cost, which means a miner's viability depends on securing both cheap machines and cheap power at the same time. Operators competing for block rewards need powerful equipment and a stable electricity supply, and because reward income is fixed per block while costs vary by location, the local power price becomes a large component of production cost. That is the half of the ledger most exposed to grid conditions: a change in the tariff a mining site pays flows almost directly into the cost of producing each coin. Miners compete for block rewards, the income stream the Bitcoin Halving periodically reduces, so the pressure to keep the other half of the bill low is structural rather than optional. AI inference runs the opposite way. Its cost is dominated almost entirely by hardware investment, with electricity taking a minor share, because an inference cluster spends its budget acquiring and maintaining computing infrastructure capable of processing large volumes of requests. Acquiring and running that infrastructure, in the comparison, accounts for most of the economics of operating an AI system. The post therefore draws a line through a debate that has grown louder as data centers and mining facilities chase the same power resources. A mining operator reads the grid price first; an AI operator reads chip supply and capacity first. Where the two industries do meet, on electricity procurement, the comparison suggests they are not bidding for the same share of their respective cost bases. Buterin offered no figures for either industry beyond the proportional split, and the post stopped at the structural observation rather than extending into forecasts for either market.

The Vantage Point Behind the Comparison

The comparison carries weight precisely because of where it comes from. Buterin speaks as the co-founder of a rival protocol, a position disclosed in the post's own framing, and readers can weigh that interest directly. Our reading of the structure is narrower: the split identifies which cost lever moves each business. Mining operators track power tariffs, since electricity is roughly half their bill and moves with grid conditions; AI operators track hardware pricing, since equipment absorbs nearly the whole of inference spending. The argument has already traveled through Bitcoin Maximalism circles, and it joins earlier coverage of the AI bubble bailout case on our desk.