The issue of Bitcoin's massive energy consumption was previously raised. However, five years after the peak of public condemnation, the situation has changed due to the development of artificial intelligence (AI).
The electricity required to secure the Bitcoin network is no longer the primary consumption metric compared to the growing demand from machines in global data centers.
Around 2021, technology publications frequently featured headlines criticizing Bitcoin for excessive energy use, comparing it to the consumption of entire countries. Nevertheless, even then, Bitcoin did not appear worse than the banking sector, which was accused of posing a threat.
A 2021 study by Galaxy Digital estimated the Bitcoin network's consumption at approximately 113 TWh per year, while the banking system consumed about 263 TWh, and gold mining consumed 240 TWh. The author notes that Galaxy is a crypto firm with an obvious interest in this answer, but emphasizes that traditional finance infrastructure also requires significant resources.
Today, the Bitcoin network consumes, according to estimates by the Cambridge Centre for Alternative Finance, about 138 TWh per year, which accounts for approximately 0.5% of global electricity, and this figure has remained relatively stable over several years.
The main misconception in older comparisons is the disregard for the global energy demand of data centers. In 2024, all global data centers consumed about 415 TWh, accounting for about 1.5% of global electricity, according to data from the International Energy Agency (IEA). Furthermore, AI is only part of this total volume.
Consumption Limitations
The IEA estimates AI's contribution to data center energy consumption in recent years to be between 5–15%, corresponding to approximately 20 TWh to 60 TWh in 2024. According to Our World in Data, global AI consumption reached about 155 TWh by 2025. In a direct comparison, AI has only begun to catch up to Bitcoin, and in 2024, Bitcoin consumed more.
However, the impact of AI is more significant. Bitcoin's consumption is limited because miners can profitably spend on earning a fixed reward, so it has barely changed over the years. AI consumption does not have such limitations. The IEA forecasts that by 2030, AI will account for 35–50% of data center electricity, amounting to 330 TWh to 470 TWh, which is two to three times the current level of Bitcoin.
The criticism against Bitcoin ceased not because Bitcoin decreased, but because attention shifted to AI data centers. Moreover, Bitcoin turns out to be a greener option. According to IEA accounting, data centers use about 27% renewable energy and 15% nuclear energy, leaving almost 60% to fossil fuels, mainly coal. Meanwhile, Bitcoin mining uses 43% renewable sources and 52% when including nuclear energy.
These differences are due to the fact that data centers must be located near populated areas and fiber optics, utilizing the local grid, whereas a miner can be located where energy is cheaper, which often coincides with places having surplus or unused clean energy. Moreover, data centers are becoming 'greener,' and the IEA expects the share of clean energy to reach 60% by 2035.
Bitcoin has become cleaner for non-emotional reasons: the miner's main goal is to minimize electricity bills, so they seek the cheapest energy on the planet, which is often surplus or unused. Cambridge estimates the network's emissions at 39.8 million tons of CO2 equivalent, corresponding to the profile of a medium industrial area, not a civilizational threat.
Stephen Boyke Sidley, a Bitcoin author from South Africa and partner at Bridge Capital, asserts that no one talks about Bitcoin's energy consumption anymore because it has become very clean, transitioning to renewables in search of cost reduction, estimating the network as 65% green. His estimate is in the optimistic range, while Cambridge's more conservative calculation is 52.4%, but the direction of movement remains unchanged.
Another important shift is related not to megawatts, but to money. The halving of the block reward in April 2024 reduced miners' income, while total computational power reached a record 1,160 exaflops in October 2025. Then, due to AI demand, DRAM contract prices sharply rose in the first quarter of 2026, nearly doubling memory costs. Although they use different chips, they compete for the same production capacity, capital, and energy.
Miners Became Landlords
The rational response was as follows: if you own the most scarce asset in AI development—a building connected to the network—why fill it with low-yield machines when an AI company will pay much more for the same megawatts? Sidley notes that the advantage of Bitcoin mining is that the miner effectively mines coins with value, part of which remains with them, unlike AI, where they are only paid for electricity.
The operator who mines retains growth potential if the coin price increases. The operator who rents out their capacity receives a fixed payment, regardless of currency fluctuations. If miners switch to AI, it does not slow down Bitcoin issuance, as issuance is tied to the number of blocks, not the number of miners. Difficulty automatically adjusts every two weeks so that blocks appear approximately every 10 minutes.
Computational power decreased by about 20% from the peak in October 2025 to 900–930 exaflops, and difficulty dropped by about 15% in 2026, but coin issuance did not lag behind schedule. Lower difficulty simply makes mining more profitable for those who remain, setting a lower profitability threshold.
Sidley warns that the real problem is not the coin supply, but network security, as lower hash rate is theoretically easier to attack. Nevertheless, at 900 exaflops, the network still significantly surpasses the capabilities of any potential attacker. Sidley calls this wide shift a great capital outflow that reoriented when the same resources became more valuable for a chatbot than for a coin.