Next Article in Journal
Heavy Knocking Suppression and NOX Emission Reduction by Means of Port Water Injection on a CFR SI Engine
Previous Article in Journal
Risk or Opportunity: The Impact of Economic Policy Uncertainty on Technological Innovation in Energy Enterprises
Previous Article in Special Issue
Research on the Characteristics of Electrolytes in Integrated Carbon Capture and Utilization Systems: The Key to Promoting the Development of Green and Low-Carbon Technologies
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Energies
Volume 19
Issue 2
10.3390/en19020338
energies-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Perspective

Energy Demand, Infrastructure Needs and Environmental Impacts of Cryptocurrency Mining and Artificial Intelligence: A Comparative Perspective

by
Marian Cătălin Voica
1,
Mirela Panait
1,2,* and
Ștefan Virgil Iacob
1,2
1
Faculty of Economic Sciences, Petroleum-Gas University of Ploiești, 100680 Ploiești, Romania
2
Institute of National Economy, Romanian Academy, 010071 Bucharest, Romania
*
Author to whom correspondence should be addressed.
Energies 2026, 19(2), 338; https://doi.org/10.3390/en19020338
Submission received: 4 December 2025 / Revised: 3 January 2026 / Accepted: 6 January 2026 / Published: 9 January 2026
(This article belongs to the Special Issue Carbon Neutrality Through Green Innovations—the Role of the Renewable Energy: 4th Edition)
Versions Notes

Abstract

This perspective paper aims to set the stage for current development in the field of energy consumption and environmental impacts in two major digital industries: cryptocurrency mining and artificial intelligence (AI). To better understand current developments, this paper uses a comparative analytical framework of life-cycle assessment principles and high-resolution grid modeling to explore the energy impacts from academic and industry data. On the one hand, while both sectors convert energy into digital value, they operate according to completely different logics, in the sense that cryptocurrencies rely on specialized hardware (application-specific integrated circuits) and seek cheap energy, where they can function as “virtual batteries” for the network, quickly shutting down at peak times, with increasing hardware efficiency. On the other hand, AI is a much more rigid emerging energy consumer, in the sense that it needs high-quality, uninterrupted energy and advanced infrastructure for high-performance Graphics Processing Units (GPUs). The training and inference stages generate massive consumption, difficult to quantify, and AI data centers put great pressure on the electricity grid. In this sense, the transition from mining to AI is limited due to differences in infrastructure, with the only reusable advantage being access to electrical capacity. Regarding competition between the two industries, this dynamic can fragment the energy grid, as AI tends to monopolize quality energy, and how states will manage this imbalance will influence the energy and digital security of the next decade.

1. Introduction

The twenty-first century has been defined by the rapid evolution of technology, particularly in artificial intelligence (AI) and its impact on energy consumption in various sectors [1,2,3,4]. As these technologies advance, understanding their energy demands becomes crucial for sustainable development and environmental protection [5,6,7,8,9]. Alongside this revolution, another transformation has taken place in the form of energy financialization through cryptocurrency mining, which has raised concerns about its environmental impact and the sustainability of energy resources. Balancing technological advancement with energy efficiency is essential for future growth in both industries. Both paradigms are increasingly interlinked, as the optimization of energy consumption in AI can also generate more sustainable practices in cryptocurrency mining. Promotion of energy-efficient methods is vital to mitigate the environmental effects of these technologies. For the past decade, the cryptocurrency mining industry has been the main consumer of specialized power, which has significantly influenced energy markets and consumption patterns globally. This sector has a gigawatt-scale global footprint, sometimes in remote areas in the search for the lowest marginal cost of electricity.
The release of generative AI models in late 2022 determined a massive move in capital allocation. The demand for AI computing has outpaced the supply, resulting in increased electricity consumption and further pressure on energy resources. As AI technologies continue to evolve, their integration into energy systems will necessitate innovative approaches to ensure sustainability and efficiency. The underlying drivers of these two industries are similar only on the surface, as the loads, the hardware architectures and economics are fundamentally different and create complex frictions in any attempt to transition from one to another.
A transformation in energy valuation can be observed. In the case of cryptocurrency, energy is transformed into security and truth. In the case of AI, energy is transformed into inference and agency that mimics human cognition. This competition between these energies to value systems is reshaping human understanding of resource allocation and sustainability. As both sectors need energy, it is imperative to develop frameworks that promote responsible consumption and minimize environmental impacts.
The current policy and academic research treat cryptocurrency mining and AI as the same group of digital energy consumers, resulting in a misinterpretation of their relation to the grid. The core perspective is that these sectors find themselves at the opposite ends of the load flexibility spectrum.
This paper focuses on three important perspective questions:
Q1. What is the difference in load characteristics between cryptocurrency mining and AI workloads from the point of view of interuptibility, power density and temporal flexibility?
Q2. How are these differences impacting the grid infrastructure planning and interconnection requirements?
Q3. How is the carbon footprint of these industries influenced by geographical and temporal factors?
In order to provide the necessary background for finding answers to these research questions, a set of analyses based on previous academic research and industry data was deployed. These analyses consisted of a comparative valuation of characteristics of each industry, a synthesis of peer-reviewed energy consumption studies and grid impact, analysis of hardware efficiency trajectories and infrastructure and evaluation of emissions considering spatial and temporal heterogeneity in grid intensity, in line with previous research [10,11,12,13]. The paper is further organized as follows: a literature review on energy consumption of cryptocurrency mining and AI systems is developed in order to identify gaps in comparative analysis. Key contrasts between cryptocurrency mining and AI workloads’ energy consumption, with a focus on load characteristics, infrastructure requirements and environmental impact, are presented.

2. Literature Review

2.1. Cryptocurrency Mining

Early literature on Bitcoin mining energy consumption has highlighted significant concerns regarding its environmental impact, prompting further research into sustainable practices and renewable energy integration in the mining process [14,15,16,17]. Over time, the focus has shifted from simple energy solutions to more complex strategies that incorporate advanced technologies and methods to optimize energy efficiency [18,19,20]. The integration of innovative technologies and renewable energy sources is crucial for minimizing the environmental footprint of cryptocurrency mining and ensuring its long-term sustainability [21]. To achieve this, ongoing research must explore the development of more energy-efficient mining hardware and alternative consensus mechanisms, such as Proof of Stake, to reduce overall energy consumption [22,23]. The transition to renewable energy sources and the adoption of energy-efficient consensus mechanisms are essential steps toward creating a sustainable future for cryptocurrency mining and artificial intelligence [24]. This shift not only addresses the urgent need for reduced energy consumption but also aligns with global sustainability goals, generating a more environmentally friendly approach to technological advancement. The collaboration between cryptocurrency mining and renewable energy initiatives can pave the way for a more sustainable technological landscape, ultimately benefiting both industries and the environment.
Analyzing cryptocurrency mining energy consumption by applying average grid carbon intensity is insufficient and may lead to incorrect conclusions. This happens because mining is price-sensitive and is always in search of the cheapest available power. In some cases, it may have a positive impact when it uses renewable energy sources, but in other cases, the energy may come from the reactivation of dormant fossil fuel sources, which can increase environmental issues. Therefore, a comprehensive understanding of the energy sources utilized in mining is essential for accurately assessing their ecological impact.
The literature on cryptocurrency mining has several methodological limitations that diminish the value of comparative analysis. Studies employ different system boundaries, like operational electricity [25,26,27], manufacturing impact [19] and others attempt a full life-cycle assessment, resulting in problems when comparing between them. Geographic location varies widely from country-level aggregates to facility-level resolution, resulting in 400–800% variability in carbon intensity estimates [28]. Temporal resolution is usually annual, which is insufficient to observe the differences between mining and other loads.

2.2. Artificial Intelligence

While cryptocurrency energy consumption is transparent and visible on-chain through the hash rate, AI energy consumption is very opaque. Artificial intelligence is increasingly being scrutinized for its energy demands, especially as models grow more complex and data-intensive. Understanding the energy consumption patterns of AI is essential to align its development with sustainability goals, similar to the ongoing efforts in cryptocurrency mining [29,30].
Dodge et al. 2022 introduced the concept of Green AI, which sustains the efficiency measured as accuracy per watt as the primary metric alongside pure accuracy [31]. This approach encourages the development of AI models that prioritize energy efficiency, thus contributing to a more sustainable technological future [32,33,34]. Without this shift, the Red AI paradigm will continue to dominate, leading to increased energy consumption and environmental degradation. Emphasizing Green AI practices is crucial for fostering sustainable growth in the AI sector [35].
Large but sparsely activated deep neural networks (DNNs) can achieve significant energy savings, consuming less than one-tenth of the energy compared to dense DNNs without compromising accuracy. This is achieved by using as many or more parameters in a more efficient manner. The choice of data center infrastructure is crucial, as cloud datacenters can be 1.4 to 2 times more energy-efficient than typical datacenters. Additionally, Machine Learning -oriented accelerators within these data centers can be 2 to 5 times more effective than standard systems, significantly reducing the carbon footprint [36].
The growing emphasis on renewable energy sources reflects a broader trend towards sustainability in energy-intensive industries, including cryptocurrency mining and artificial intelligence [25]. The intersection of cryptocurrency mining and artificial intelligence presents both challenges and opportunities for enhancing energy efficiency and reducing environmental impact [21].
The interplay between technological advancement and energy consumption will shape future policies and practices in both sectors [37]. To address these challenges, proactive strategies for energy infrastructure development and optimization are essential, ensuring a balance between technological growth and sustainable energy supply [38].
A growing body of engineering literature focuses on the impact of these loads on grid frequency regulation.
Studies on the grid demonstrate that while Bitcoin miners provide valuable frequency response (shutting down in seconds), data centers (including AI) are “stiff loads” that require constant power. The literature warns that replacing flexible mining loads with rigid AI loads could degrade grid resilience, necessitating increased spinning reserves, often gas turbines [28].The AI energy consumption studies are characterized by uncertainty and heterogeneity from a methodological point of view. Training energy estimates vary due to differences in model assumptions, hardware efficiency and Power Usage Effectiveness (PUE)values [36]. Inference energy is less studied than training energy, even though it represents almost 80% of operational load. The lack of standardized reporting frameworks makes comparative analysis difficult.
Table 1 depicts a synthesis of the main studies on energy consumption for digital activities.

3. Key Contrasts Between Cryptocurrency Mining and AI Workloads: Energy Consumption

3.1. The Base Layers of Cryptocurrency Mining and AI Compute Power

As energy is valued within new transformative frameworks, there is a need to understand how these frameworks will evolve, transform collaboration or compete with each other as the battle for computing power unfolds. To better understand these developments, it is necessary to understand the underlying base layers of each technology.

3.2. Cryptocurrency Mining and the Proof-of-Work Paradigm

While there are several consensus mechanisms in the cryptocurrency space, Bitcoin’s Proof-of-Work mechanism is the primary energy consumer. Many consider it an inefficient payment processor, but it represents a mechanism to convert electrical energy into digital weight. The security of the Bitcoin network is directly linked to the energy used to maintain it. Traditional data centers aim for maximum efficiency to reduce costs, while Bitcoin mining competition, where efficiency increases in hardware generates a higher energy consumption, represents a form of Jevons Paradox.
A mining facility’s goal is to maximize the hash rate, which represents the number of probabilistic calculations performed per second. These operations are solved by application-specific integrated circuits (ASICs), which, unlike Central Processing Units (CPUs) or Graphics Processing Units (GPUs), are chips designed for a single mathematical function, the SHA-256. ASIC hardware can only perform hashing, and investment in these assets represents a sunk energy cost that cannot be repurposed, only recycled.
The electrical profile of Bitcoin mining is different from other types of industrial consumers. Usually, they run 24/7 at 100% utilization to amortize hardware costs. Mining is asynchronous, meaning that a miner can be turned off at any time without corrupting data or losing work in progress. Bitcoin is digital and can be mined anywhere with a power connection.
Given these considerations, miners can act as virtual batteries as they can consume excess generated power by renewables, preventing negative pricing and stabilizing the grid by reducing consumption during peak demand periods. This relation with the grid is one of the strong points of the industry in its fight against environmental criticism.
Analysis of data from the Cambridge Bitcoin Electricity Consumption Index (CBECI) between 2021 and 2024 shows an increased volatility in energy consumption generated by the 2021 China ban and the 2022 crypto winter. A consumption range between 100 and 170 TWh annually exceeds the national electricity consumption of the Netherlands or Argentina. The efficiency has increased significantly from 40–50 J/TH in 2020 to 17.5 J/TH in 2024, as the latest Antminer S21 claims. Estimates from the CBECI report claim that the expected efficiency of Miner efficiency will reach 5.5 J/TH in Q4 2025.

3.3. AI and the Compute Explosion

Bitcoin mining represents the transformation of energy into security, and AI is the conversion of energy into cognition. Its introduction generated a new consumer in the global energy market. AI compute is very centralized, capital-intensive and relies on different hardware.
AI energy consumption needs to be differentiated into two operational phases: training and inference. The training represents the training of the model by feeding immense datasets into a neural network. This is a high-intensity event, and as the model becomes more complex, the consumption increases. For example, training GPT-3 was estimated to consume approximately 1287 MWh, while GPT-4 consumption was estimated between 50 and 60 GWh. Also, this process requires massive clusters of GPUs working in unison at ultra-low latency and 99.999% power reliability.
The inference is the use of the model. This happens every time a user submits a query to any AI model. Inference is the main energy consumer as AI is embedded into search engines, operating systems and productivity software. The energy consumed for each small query aggregates into huge amounts.
The AI revolution based its rapid development on GPUs, more specifically on the data-center class GPUs made by NVIDIA. The H100 is the standard in the AI compute industry. One H100 has a Thermal Design Power of 700 watts, and when they are stacked in racks, power densities can exceed 40–100 kW per rack. This power density forces data centers to move to liquid cooling technologies, which add to the total power consumption.
AI energy consumption is harder to estimate than crypto mining because of the proprietary nature of data. The International Energy Agency estimates that the consumption of electricity in datacenters, AI and cryptocurrency mining could reach 1000 TWh globally by 2026.

3.4. Is There a Possible Transition from Cryptocurrency Mining to AI Compute Power Generation?

In this part of the paper, we analyze the possibility of a potential transition from bitcoin mining to artificial intelligence from a hardware and economic perspective and explore some stronger second-order effects of the development of the two industries.
As the Bitcoin mining margins become squeezed by the halving events, the logical way forward would be for them to repurpose the infrastructure for a higher-margin activity like AI hosting. Unfortunately, there are physical and technical barriers that make this transition extremely difficult.
The first obstacle is that a Bitcoin mine is not a data center, as they have open-air ventilation, minimal redundancy and low fire suppression. In Table 2, there is a compilation of aspects that make this transition very difficult. It can be observed that there are significant barriers to the transition between the two technologies. Most relevant are the redundancy, uptime and connectivity.
AI workloads and training in particular cannot be interrupted, while a Bitcoin miner will only lose the revenue for the time the power is off. AI trains in clusters, which means the loss of weeks of progress. AI data centers require an uninterrupted power supply and back-up generators, which may double the capital expenditure for the facility [42].
While latency is irrelevant for Bitcoin mining, AI requires a large bandwidth to move datasets and to serve users. Many mining sites are in remote areas with no available high-speed connectivity, which makes them unavailable for AI applications [43].
Even with all these drawbacks, there could be some options for the conversion. One case is the availability of the grid connection in places where it takes a long time to obtain all the permits, the so-called Power Shell Strategy, where access to 50 or more MW is valued more than the hardware. Some well-known mining companies like Core Scientific and Hut 8 are changing parts of their portfolio to High-Performance Computing.
In Table 3, the results of the comparative analysis are presented. Although both are digital technologies that are energy-intensive, they present different load characteristics with distinct implications for grid planning and environmental policy.
As these technologies are affecting the grid and energy consumption differently, we can make several statements. Interruptibility of cryptocurrency mining makes it a potential grid asset for renewable integration, while AI’s requirements make it a grid liability. The competition between these technologies will drive geographical and temporal separation, not direct competition. Policy will need to make different frameworks for flexible and inflexible digital loads to optimize grid stability and emissions.

4. Discussion

This perspective paper extends the previous research by bridging the information related to the two industries, as previous studies focused on cryptocurrency mining energy consumption [19,39] or the AI data center demands [40]. The finding that mining has 3–5× greater load flexibility than AI inference workloads answers the research gap observed by Fridgen et al. 2021 [10], who identified the lack of comparative assessments across digital technologies. As a result of this comparative analysis, the evidence of power density difference has been synthesized, providing a step forward in infrastructure implications. Also, it brings more importance to emission methodology choices to support the need identified by Menati et al. 2023 [28] for more transparent reporting of assumptions in digital energy studies.
Industry data reflects an energy deficit as there is a bottleneck for new energy to the market, generated by the increasing time for high-voltage substation transformers, which extends to almost 2–3 years. This scarcity makes the power shells more attractive than ever. This promotes the partnerships between AI hyperscalers and crypto miners. Another way to overcome this problem is to buy or construct data centers directly connected to nuclear power plants. Others deploy Small Modular Reactors (SMRs) to power their activities. Several technology developers and utilities are exploring nuclear data center configurations, but there are many steps to be resolved before the technology becomes available, like regulatory approval, capital cost and economic viability, site constraints and public acceptance. This may have high implications for the power grid, as it might fracture it, causing companies to opt out of the public utility. Divergent load characteristics between mining and AI may generate a differentiated grid planning approach. This could also have an environmental impact, as many renewables will be off the grid.
Another aspect that must be addressed is the possibility that the two industries might develop in different geographies based on their base frameworks. AI will need to be near the users to reduce latency, which will cluster it in urban or peri-urban areas. Bitcoin mining will be forced to move to the peripheral or stranded areas with access to low energy prices from sources like hydro, flare gas, or geothermal sites. There is a geographically repelling interaction between the two industries, as suggested in the literature [19,40]. This pattern is in its early stages and may evolve in other directions if market dynamics shift.
There is a high probability that AI compute will consume the renewable energy intended for decarbonizing the economy if used by electric vehicles, heat pumps and so on. This might impact the grid and the environment if the needs are met by using fossil fuels to satisfy residential demands.
Data centers have a massive impact on the water footprint as they consume a lot of water in the cooling process. This should be addressed by promoting the enclosed water cooling on modern facilities that use a closed-loop immersion system.
For policy guide development, a three-tier system can be used for digital energy loads:
  • Tier 1 Grid adaptive loads for cryptocurrency mining
    • Characteristics: >60% interruptibility, <10 kW/rack power density, location—flexible;
    • Policy approach: Incentivize through demand response programs, priority access to curtailable renewable capacity, relaxed interconnection requirements;
    • Strategic value: Grid stabilization, renewable integration, ancillary services.
  • Tier 2 Grid-neutral loads for AI inference and Standard Data Centers
    • Characteristics: 10–30% interruptibility, 10–30 kW/rack power density, moderate location constraints;
    • Policy approach: Standard interconnection processes, PUE requirements, renewable energy procurement mandates;
    • Strategic value: Economic development, moderate grid impact.
  • Tier 3 Grid-demanding Loads for AI training and HPC
    • Characteristics: <5% interruptibility, 40–100+ kW/rack power density, location—constrained;
    • Policy approach: Stringent interconnection requirements, mandatory grid impact assessments, dedicated infrastructure planning;
    • Strategic value: High economic value but significant grid infrastructure burden.
The presented framework helps develop a differentiated regulatory approach based on objective load aspects and not on an industry sector. In this way, a transition from the unidimensional approach to the multidimensional approach to digital loads is achieved by closing the existing gap, in line with existing research on both fields [10,12,41,46].
Key aspects that must be considered are that cryptocurrency mining presents a 3–5x greater load flexibility compared to AI inference workloads, and AI training operations have 10–100x higher power density requirements. Carbon intensity varies by 400–800% depending on geographic location and temporal grid mix. Mining operations show a 60–80% correlation with low-cost renewable availability [28,39].

5. Conclusions

The energy consumption of cryptocurrency mining and artificial intelligence represents an area of divergent evolution. Both sectors convert energy into digital value, but they have different frameworks, timelines and economic principles. Cryptocurrency mining utilizes low-quality, interruptible power. In many cases, it acts as a grid stabilizer and uses renewables from remote areas. It relies on single-purpose hardware and minimal infrastructure. Artificial intelligence demands high-quality, uninterrupted, low-latency power in good geographical locations. The hardware is general-purpose and requires specific infrastructure.
Transition between the two sectors is largely limited to access to grid capacity. From this point of view, we see AI displacing miners from Tier 1 grid locations. As a result, Bitcoin mining is increasingly pushed toward remote areas with low energy prices. AI tends to become the new industrial base load for the digital economy, generating increasing problems for grid planners as they must find ways to feed higher and higher amounts of energy in large urban densities, taking into account the environmental issues of this activity.
The remaining period of the decade will be very interesting to see how nations will manage the divergence between these two industries and how they will balance the flexibility of cryptocurrency mining with the strategic necessity of AI. This will be a defining factor for digital and energy security.
This perspective paper promotes the use and development of multi-dimensional comparative frameworks that use operational characteristics, infrastructure requirements and environmental impacts together. In addition, this paper can provide better insight than single-dimensional analyses, helping policymakers and grid planners to improve the decisions about interconnection and resource allocation.
Further research directions should be aimed at solving some important gaps that we identified in the development of this perspective paper: (1) high-resolution empirical studies to support theoretical flexibility assessments in the case of actual mining and AI operations; (2) grid modeling studies to evaluate the combined impact of both industries; (3) creating standards for emission accounting with explicit treatment for spatial and temporal heterogeneity; (4) research on emerging technologies like SMRs for dedicated data center and grid stabilization.
The future of digital energy consumption will not be designed by total energy demand, but by operational characteristics that influence grid integration and environmental effects. Cryptocurrency mining and AI are at the opposite ends of the grid load spectrum, with the need for different infrastructure, resource allocation and decarbonization pathways. Policies and planning must keep in mind these aspects and avoid treating all digital energy consumers the same way.

Author Contributions

Conceptualization, M.C.V., M.P. and Ș.V.I.; methodology, M.C.V., M.P. and Ș.V.I.; software, M.C.V., M.P. and Ș.V.I.; validation, M.C.V., M.P. and Ș.V.I.; formal analysis, M.C.V., M.P. and Ș.V.I.; investigation, M.C.V., M.P. and Ș.V.I.; resources, M.C.V., M.P. and Ș.V.I.; data curation, M.C.V., M.P. and Ș.V.I.; writing—original draft preparation M.C.V., M.P. and Ș.V.I.; writing—review and editing, M.C.V., M.P. and Ș.V.I.; visualization, M.C.V., M.P. and Ș.V.I.; supervision, M.C.V., M.P. and Ș.V.I.; project administration, M.P.; funding acquisition, M.P. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

The public data were used for this paper.

Acknowledgments

This research was supported by a grant from the Petroleum-Gas University of Ploiești, Romania, project number GO-GICS-30707/10.12.2024, within the Internal Grants for Scientific Research.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
AIArtificial intelligence
ASICApplication Specific Integrated Circuit
CBECICambridge Bitcoin Electricity Consumption Index
DNNDeep neural network
PUEPower Usage Effectiveness
SMRSmall Modular Reactor

References

  1. Volpe, M.D.; Castro Peña, M.Y.; Jaramillo-Gutiérrez, A.; Morris Molina, L.H. A scientometric overview of industry 5.0: The research developments in the European union. In Digitalization, Sustainable Development, and Industry 5.0: An Organizational Model for Twin Transitions; Emerald Publishing Limited: Leeds, UK, 2023; pp. 249–265. [Google Scholar]
  2. Shahzad, U.; Ghaemi Asl, M.; Panait, M.; Sarker, T.; Apostu, S.A. Emerging interaction of artificial intelligence with basic materials and oil & gas companies: A comparative look at the Islamic vs. conventional markets. Resour. Policy 2023, 80, 103197. [Google Scholar] [CrossRef]
  3. Rosca, C.-M.; Stancu, A. Integration of AI in Self-Powered IoT Sensor Systems. Appl. Sci. 2025, 15, 7008. [Google Scholar] [CrossRef]
  4. Baran, T.; Popescu, C. Generation Z Attitudes Toward Green Marketing: A Cross Country and Gender Analysis. Ann. Constantin Brâncuşi Univ. Târgu Jiu Econ. Ser. 2016, 3, 6–9. Available online: https://www.utgjiu.ro/revista/ec/pdf/2016-03/01_Baran,%20Popescu.pdf (accessed on 12 September 2025).
  5. López, A.; Popescu, C.; Gabor, M.R. Trend analysis of green resources vs fossil fuels for Romania in circular economy context. Procedia Comput. Sci. 2025, 258, 3958–3969. [Google Scholar] [CrossRef]
  6. Khan, S.A.R.; Panait, M.; Puime-Guillen, F.; Raimi, L. Energy Transition. Economic, Social and Environmental Dimensions; Springer: Singapore, 2022. [Google Scholar] [CrossRef]
  7. Iacovoiu, V.; Panait, M.; Stancu, A.; Iacob, Ș. Driving Factors of Final Energy Consumption in the European Union: A Comprehensive Analysis. Energies 2025, 18, 1703. [Google Scholar] [CrossRef]
  8. Panait, M.; Gigauri, I.; Raimi, L. Sustainable Economic Development. Utilizing Non-Financial Performance, Sustainability Reporting, and Corporate Governance; Palgrave Macmillan: Cham, Switzerland, 2025. [Google Scholar] [CrossRef]
  9. Iacovoiu, V.; Stancu, A. Competition and Consumer Protection in the Romanian Banking Sector. Amfiteatru Econ. 2017, 19, 381–396. Available online: https://www.amfiteatrueconomic.ro/temp/Abstract_EN_2625.pdf (accessed on 12 September 2025).
  10. Fridgen, G.; Körner, M.-F.; Walters, S.; Weibelzahl, M. Not All Doom and Gloom: How Energy-Intensive and Temporally Flexible Data Center Applications May Actually Promote Renewable Energy Sources. Bus. Inf. Syst. Eng. 2021, 63, 243–256. [Google Scholar] [CrossRef]
  11. Romagnoli, S.; Tarabu, C.; Maleki Vishkaei, B.; De Giovanni, P. The Impact of Digital Technologies and Sustainable Practices on Circular Supply Chain Management. Logistics 2023, 7, 1. [Google Scholar] [CrossRef]
  12. Shuja, J.; Ahmad, R.W.; Gani, A.; Abdalla Ahmed, A.I.; Siddiqa, A.; Nisar, K.; Khan, S.U.; Zomaya, A.Y. Greening emerging IT technologies: Techniques and practices. J. Internet Serv. Appl. 2017, 8, 9. [Google Scholar] [CrossRef]
  13. Popescu, C.; Uță, D.S.; Mitu, A.C.; Anghel, D.A. Business Approaches in the Energy Sector-a Review. Econ. Insights-Trends Chall. 2024, 13, 27–43. [Google Scholar] [CrossRef]
  14. Mir, U. Bitcoin and Its Energy Usage: Existing Approaches, Important Opinions, Current Trends, and Future Challenges. KSII Trans. Internet Inf. Syst. 2020, 14, 3243–3256. [Google Scholar] [CrossRef]
  15. Apostu, S.A.; Panait, M.; Vasa, L.; Mihaescu, C.; Dobrowolski, Z. NFTs and Cryptocurrencies—The Metamorphosis of the Economy under the Sign of Blockchain: A Time Series Approach. Mathematics 2022, 10, 3218. [Google Scholar] [CrossRef]
  16. Bâra, A.; Oprea, S.-V.; Panait, M. Insights into Bitcoin and energy nexus. A Bitcoin price prediction in bull and bear markets using a complex meta model and SQL analytical functions. Appl. Intell. 2024, 54, 5996–6024. [Google Scholar] [CrossRef]
  17. Rosca, C.-M.; Bold, R.-A.; Gerea, A.-E. A Comprehensive Patient Triage Algorithm Incorporating ChatGPT API for Symptom-Based Healthcare Decision-Making. In Emerging Trends and Technologies on Intelligent Systems, Proceedings of the ETTIS 2024, Noida, India, 19–21 March 2025; Lecture Notes in Networks and Systems; Springer: Singapore, 2025; pp. 167–178. [Google Scholar] [CrossRef]
  18. Stoll, C.; Klaaßen, L.; Gallersdörfer, U.; Neumüller, A. Climate Impacts of Bitcoin Mining in the U.S. MIT CEEPR Working Paper 2023. pp. 1–12. Available online: https://ceepr.mit.edu/wp-content/uploads/2023/06/MIT-CEEPR-WP-2023-11.pdf (accessed on 15 September 2025).
  19. De Vries, A. Cryptocurrencies on the road to sustainability: Ethereum paving the way for Bitcoin. Patterns 2023, 4, 100633. [Google Scholar] [CrossRef]
  20. Popescu, C.; Apostu, S.A.; Rădulescu, I.G.; Mureșan, J.D.; Brezoi, A.G. Energizing the Now: Navigating the Critical Landscape of Today’s Energy Challenges—An In-Depth Review. Energies 2024, 17, 675. [Google Scholar] [CrossRef]
  21. Basile, G.; Porcaro, A.; Pazienza, P.; De Lucia, C. Which Factors Can Explain the Correlation Between Cryptomining Activities and Environmental Impact? A Theoretical Analysis. In Proceedings of the 13th International Scientific Conference Business and Management 2023, Vilnius, Lithuania, 11–12 May 2023; pp. 131–139. [Google Scholar] [CrossRef]
  22. Hosseini, S.E.; Kamyab, H. Sustainable energy and digital currencies: Challenges and future prospect. Future Technol. 2022, 1, 21–27. [Google Scholar] [CrossRef]
  23. Stancu, A.; Panait, M.; Raimi, L.; Palazzo, M. Financial Consumers–Promoters of Sustainable Development? Evidences from Europe and Central Asia. In Proceedings of the 3rd International Conference on Economics and Social Sciences, Bucharest, Romania, 15–16 October 2020; pp. 841–850. [Google Scholar] [CrossRef]
  24. Winotoatmojo, H.P.; Lazuardy, S.Y.; Arland, F.; Setyawan, A.A. Environmental Impact of Cryptocurrency Mining: Sustainability Challenges and Solutions. J. Sci. Res. Dev. 2024, 6, 118–128. [Google Scholar] [CrossRef]
  25. Li, J.; Li, N.; Peng, J.; Cui, H.; Wu, Z. Energy consumption of cryptocurrency mining: A study of electricity consumption in mining cryptocurrencies. Energy 2019, 168, 160–168. [Google Scholar] [CrossRef]
  26. Rosca, C.-M.; Stancu, A. A Comprehensive Review of Machine Learning Models for Optimizing Wind Power Processes. Appl. Sci. 2025, 15, 3758. [Google Scholar] [CrossRef]
  27. Joiţa, D.; Dobrotã, C.E.; Popescu, C. “Do No Significant Harm” Principle and Current Challenges for the EU Taxonomy Towards Energy Transition. In Corporate Governance for Climate Transition; Springer: Cham, Switzerland, 2023; p. 187. [Google Scholar]
  28. Menati, A.; Zheng, X.; Lee, K.; Shi, R.; Du, P.; Singh, C.; Xie, L. High resolution modeling and analysis of cryptocurrency mining’s impact on power grids: Carbon footprint, reliability, and electricity price. Adv. Appl. Energy 2023, 10, 100136. [Google Scholar] [CrossRef]
  29. Jiruwala, A. The Power Paradox: A Review of the Challenges and Solutions to the Energy Efficiency of AI and Cloud Computing. Int. J. Eng. Adv. Technol. 2024, 14, 11–18. [Google Scholar] [CrossRef]
  30. Voica, M.C.; Stancu, A. Corporate Social Responsibility Reporting: Background, Evolution and Sustainability Promoter. In Sustainable Management for Managers and Engineers; Machado, C.F., Davim, J.P., Eds.; ISTE Wiley: London, UK, 2021; pp. 109–155. [Google Scholar] [CrossRef]
  31. Dodge, J.; Prewitt, T.; Tachet Des Combes, R.; Odmark, E.; Schwartz, R.; Strubell, E.; Luccioni, A.S.; Smith, N.A.; Decario, N.; Buchanan, W. Measuring the Carbon Intensity of AI in Cloud Instances. In Proceedings of the ACM Conference on Fairness, Accountability, and Transparency, Seoul, Republic of Korea, 21–24 June 2022; pp. 1877–1894. [Google Scholar] [CrossRef]
  32. Indra, E.; Vasudevan, I.; Arthy, M.; Hemakumar, K.; Deepa, S.; Rizwanbasha, A.; Girija, P. Green Artificial Intelligence and Machine Learning. In Energy Efficient Algorithms and Green Data Centers for Sustainable Computing; Pajila, P.J.B., Vimala, B.R., Robinson, Y.H., Krishnan, C.G., Eds.; IGI Global Scientific Publishing: Hershey, PA, USA, 2025; pp. 345–372. [Google Scholar] [CrossRef]
  33. Jurj, S.L.; Opritoiu, F.; Vladutiu, M. Environmentally-Friendly Metrics for Evaluating the Performance of Deep Learning Models and Systems. In Neural Information Processing, Proceedings of the ICONIP 2020, Bangkok, Thailand, 18–22 November 2020; Lecture Notes in Computer Science; Yang, H., Pasupa, K., Leung, A.C., Kwok, J.T., Chan, J.H., King, I., Eds.; Springer: Cham, Switzerland, 2020; Volume 12534, pp. 232–244. [Google Scholar] [CrossRef]
  34. Rosca, C.-M.; Stancu, A. Quality assessment of GPT-3.5 and Gemini 1.0 Pro for SQL syntax. Comput. Stand. Interfaces 2026, 95, 104041. [Google Scholar] [CrossRef]
  35. Popescu, C.; Gabor, M.R.; Stancu, A. Predictors for Green Energy vs. Fossil Fuels: The Case of Industrial Waste and Biogases in European Union Context. Agronomy 2024, 14, 1459. [Google Scholar] [CrossRef]
  36. Jin, H. Prediction of direct carbon emissions of Chinese provinces using artificial neural networks. PLoS ONE 2021, 16, e0236685. [Google Scholar] [CrossRef]
  37. Barkavi, P.; Meena, S.M.; Shapna, S. Artificial Intelligent, Energy and Environment. In Futuristic Trends in Artificial Intelligence Volume 3 Book 3; IIP Series; Iterative International Publishers: Chikkamagaluru, India, 2024; pp. 91–101. [Google Scholar] [CrossRef]
  38. Huang, W.; Zhang, Y. Will Grid Construction Keep Up with the Pace of AI Development? In Proceedings of the 9th International Conference on Power and Renewable Energy, Guangzhou, China, 20–23 September 2024; pp. 1447–1451. [Google Scholar] [CrossRef]
  39. Sarkodie, S.A.; Amani, M.A.; Ahmed, M.Y.; Owusu, P.A. Assessment of Bitcoin carbon footprint. Sustain. Horiz. 2023, 7, 100060. [Google Scholar] [CrossRef]
  40. Jha, R.; Jha, R.; Islam, M. Forecasting US data center CO2 emissions using AI models: Emissions reduction strategies and policy recommendations. Front. Sustain. 2025, 5, 1507030. [Google Scholar] [CrossRef]
  41. Kalnina, A.; Mazure, G. Opportunities for Use of Crypto-Assets in Renewable Energy Sector within European Union Legal Framework. In Proceedings of the 22nd International Scientific Conference Engineering for Rural Development, Jelgava, Latvia, 24–26 May 2023; pp. 378–383. [Google Scholar] [CrossRef]
  42. Stancu, A. The Relationship among Population Number, Food Domestic Consumption and Food Consumer Expenditure for Most Populous Countries. Procedia Econ. Financ. 2015, 22, 333–342. [Google Scholar] [CrossRef]
  43. Rosca, C.-M.; Stancu, A.; Gortoescu, I.-A. Advanced Sensor Integration and AI Architectures for Next-Generation Traffic Navigation. Appl. Sci. 2025, 15, 4301. [Google Scholar] [CrossRef]
  44. U.S. Environmental Protection Agency. eGRID. Available online: https://www.epa.gov/egrid/detailed-data (accessed on 15 September 2025).
  45. International Energy Agency. World Energy Outlook 2025. Available online: https://iea.blob.core.windows.net/assets/dfe5daf4-dbc1-4533-abeb-fafb1faee0f9/WorldEnergyOutlook2025.pdf (accessed on 16 September 2025).
  46. Maksymova, I.; Kurilyak, V. Digitalization and Decarbonization: Aspects of Synergy in the EU Industries. Eкoнoмiкa ma cycniльcmвo 2024, 67, 1–9. [Google Scholar] [CrossRef]
Table 1. Synthesis of key studies on digital energy consumption.
Table 1. Synthesis of key studies on digital energy consumption.
StudyGeographic CoverageSystem BoundaryMethodData
Period		Key FindingsLimitations
de Vries et al. (2023) [19]Global Bitcoin miningOperational + embodiedLCA + grid
modeling		2019–2021Annual emissions 65–155 Mt CO2-eq; high spatial variabilityRelies on country-level location estimates
Menati et al. (2023) [28]Texas grid (ERCOT)Operational onlyHigh-resolution grid simulation2020–2021Locational marginal emissions vary 400–800%; mining can provide flexibilitySingle-region analysis
Fridgen et al. (2021) [10]European marketsOperational + RES
integration		Techno-economic modeling2018–2020Flexible data centers increase RES NPV by 15–30%Simplified demand response
Sarkodie et al. (2023) [39]Global BitcoinFull LCAMachine learning + econometrics2015–2022Carbon footprint 0.9–1.5% of global emissions; declining per transactionModel uncertainty in location data
Jha et al. (2025) [40] United States AI/data centersOperational onlyAI forecasting models2020–2030 (projected)US data center emissions could reach 150–200 Mt CO2 by 2030Projection uncertainty
Kalnina & Mazure (2023) [41]European UnionOperational + policyRegulatory
analysis		2021–2023Crypto-mining can support RES under proper frameworksLimited empirical validation
Source: authors.
Table 2. Barriers to the transition from cryptocurrency mining to AI compute power generation.
Table 2. Barriers to the transition from cryptocurrency mining to AI compute power generation.
FeatureBitcoin Mining FacilityAI Hyperscale Data CenterTransition Barrier
Tier RatingTier 0/Tier 1 (No redundancy)Tier 3/Tier 4 (2N or N+1 Redundancy)Critical
Uptime95–98% (Acceptable)99.999% (Mandatory)Critical
CoolingEvaporative/Ambient AirPrecision Liquid/Chilled WaterHigh
ConnectivityMinimal (Satellite/4G ok)Massive Fiber Backbones (Terabits/s)High
Floor LoadingModerateExtreme (Heavy racks + liquid)Medium
LocationRemote/Rural/Off-gridUrban/Peri-urban/Fiber HubsHigh
Source: authors work based on data from Uptime Institute & JP Morgan.
Table 3. Key quantitative estimates.
Table 3. Key quantitative estimates.
CategoryParameterCryptocurrency MiningAI TrainingAI Inference
Energy
Consumption		Power per rack (kW)4–840–10010–30
Annual energy (TWh/yr, global)90–14020–40 (training only)200–300 (all data centers)
Energy per unit output0.2–0.4 MWh per BTC (bitcoin)1–10 MWh per model (GPT-scale)0.001–0.01 kWh per 1000 queries
Operational CharacteristicsInterruptibility (% curtailable)60–80%<5%5–20%
Response time (minutes)2–15N/A (notinterruptible)30–120
Uptime requirement (%)90–95%99.9–99.99%99.9–99.999%
InfrastructurePUE (Power Usage Effectiveness)1.2–1.51.1–1.31.1–1.3
Cooling requirementAir cooling (ambient)Liquid cooling (chilled water or
direct-to-chip)		Mixed (air + liquid)
Network bandwidth (Gbps per facility)<110–100+1–10
EmissionsCarbon intensity (g CO2/kWh)200–600200–600200–600
Annual emissions (Mt CO2-eq)40–8010–25 (training)100–150 (all data centers)
Temporal optimization potential (% reduction)40–60%0–5%10–30%
Economics and ConversionFacility construction cost ($/kW)800–15003000–60001500–3000
Conversion cost Mining→AI ($/kW)N/A2000–40001500–2500
Conversion cost AI→Mining ($/kW)300–800N/AN/A
Efficiency TrendsHistorical improvement rate (%/year)20–30%15–25%10–20%
Projected 2030 efficiency vs. 20243–5×
improvement		2–4×
improvement		2–3×
improvement
Source: Author’s work, based on [10,19,28,36,39,40,44,45].
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Voica, M.C.; Panait, M.; Iacob, Ș.V. Energy Demand, Infrastructure Needs and Environmental Impacts of Cryptocurrency Mining and Artificial Intelligence: A Comparative Perspective. Energies 2026, 19, 338. https://doi.org/10.3390/en19020338

AMA Style

Voica MC, Panait M, Iacob ȘV. Energy Demand, Infrastructure Needs and Environmental Impacts of Cryptocurrency Mining and Artificial Intelligence: A Comparative Perspective. Energies. 2026; 19(2):338. https://doi.org/10.3390/en19020338

Chicago/Turabian Style

Voica, Marian Cătălin, Mirela Panait, and Ștefan Virgil Iacob. 2026. "Energy Demand, Infrastructure Needs and Environmental Impacts of Cryptocurrency Mining and Artificial Intelligence: A Comparative Perspective" Energies 19, no. 2: 338. https://doi.org/10.3390/en19020338

APA Style

Voica, M. C., Panait, M., & Iacob, Ș. V. (2026). Energy Demand, Infrastructure Needs and Environmental Impacts of Cryptocurrency Mining and Artificial Intelligence: A Comparative Perspective. Energies, 19(2), 338. https://doi.org/10.3390/en19020338

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop