Demand for AI and Green Skills

• Demand for AI and green skills is defined by the integration of advanced machine learning techniques with sustainability strategies to mitigate environmental impact.
• Structured evaluation methods, including carbon and energy metrics, enable transparent benchmarking of AI system efficiency across its lifecycle.
• Economic trends show that combining AI expertise with green skills yields higher wages and broader labor opportunities in tech-driven, eco-conscious sectors.

The demand for AI and green skills is driven by rapid technological advancement, the increasing environmental impact of AI, and evolving workforce needs across sectors. AI expertise now extends beyond technical proficiency in machine learning and software engineering to encompass energy efficiency, environmental sustainability, and interdisciplinary competencies—collectively referred to as green skills. This convergence reflects both the opportunities AI offers for decarbonization and operational optimization, as well as the imperative to mitigate the ecological footprint of AI systems throughout their lifecycle, from model training to deployment.

1. Environmental Impact and Challenges of AI Systems

The growth of AI, particularly deep learning, has resulted in a dramatic rise in computational demand and energy consumption, leading to significant environmental ramifications. For example, computational requirements in deep learning have increased by a factor of 300,000 from 2012 to 2018 (Schwartz et al., 2019). Large-scale models such as BERT-large and Grover were trained on 64 to 256 TPUs for several days or weeks, incurring both financial costs (e.g., \$25,000 for a single training run) and considerable carbon emissions (Schwartz et al., 2019). The carbon footprint of training a single NLP model can be comparable to that of five cars over their lifetime or hundreds of transcontinental flights (Pachot et al., 2022). With data center electricity use having doubled between 2015 (under 200 TWh) and 2024 (over 400 TWh), and projections as high as 1700 TWh for data center consumption by 2035, AI workloads are a dominant contributor to this trend (Devarakota et al., 14 Jul 2025). Beyond operational energy use, the manufacture of AI hardware incurs additional embedded emissions, with sophisticated servers exhibiting manufacturing footprints of over 1,000 kg CO₂e and annual operational footprints easily surpassing 1,500 kg CO₂e depending on region and energy mix (Clemm et al., 1 May 2024).

Despite the originally energy-efficient inspiration from the human brain, current AI systems are far less efficient, and efforts to increase model efficiency have resulted in a rebound effect: improved efficiency often enables larger and more complex models, counterbalancing or even increasing total energy consumption (Pachot et al., 2022). This highlights the need for holistic approaches that assess environmental impact over the full system lifecycle, including both hardware and software.

2. Methodologies for Sustainable AI: Metrics, Assessment, and Ecodesign

Structured methods for assessing and optimizing the environmental impact of AI are increasingly emphasized. Energy, carbon, and resource use should be integrated into evaluation criteria alongside traditional performance metrics such as accuracy and latency (Schwartz et al., 2019, Bone et al., 2023, Clemm et al., 1 May 2024, Cruz et al., 2 Jun 2025). Commonly used formulas include:

• Cost of AI Result: $Cost(R) \propto E \cdot D \cdot H$ with $E$ (cost per example), $D$ (dataset size), $H$ (hyperparameter experiments) (Schwartz et al., 2019).
• Data Center Energy: $$E_\text{total} = \text{PUE} \times e_\text{hardware}$$, with PUE = Power Usage Effectiveness (Zhou et al., 2023).
• Carbon Emissions: $\text{CO}_2 = \text{CI} \times E_\text{total}$ where CI is carbon intensity (Zhou et al., 2023, Clemm et al., 1 May 2024).
• Lifecycle Energy Footprint: $$E = (\text{Power}_\text{active} \times t_\text{active}) + (\text{Power}_\text{idle} \times t_\text{idle})$$ (Clemm et al., 1 May 2024).

Systematic benchmarking and reporting of computational cost (FLOPs, energy) and amortized metrics (e.g., $$E_\text{amortized} = E_\text{training} / N_\text{inferences}$$) are advocated to enable fair and transparent comparison (Cruz et al., 2 Jun 2025). Software engineering (SE) plays a central role in creating sustainability-aware architectures, standardizing energy measurement, and fostering modular, observable, and adaptable MLOps practices (Cruz et al., 2 Jun 2025). There is a distinct need for integrated benchmarking frameworks (e.g., MLPerf), meta-methodologies accommodating hardware and workload heterogeneity, and a taxonomy of metrics (direct, proxy, composite) (Cruz et al., 2 Jun 2025).

3. Economic Value, Labor Market Trends, and Skill Complementarity

AI and green skills derive amplified economic value from their complementarity with other high-demand competencies (Stephany et al., 2022). Wage premiums for AI skills are substantial—averaging a 21–23% premium, exceeding all but the highest academic degrees (Stephany et al., 2022, Bone et al., 2023). The most valuable profiles combine AI expertise with a diverse network of complementary skills such as data analytics, domain knowledge, or sustainability, and workers possessing such combinations are positioned for higher wages and increased employability (Stephany et al., 2022).

The labor market shows a shift towards skill-based hiring: postings for AI roles are less likely to require formal degrees and more likely to list explicit skills, averaging 15 distinct requirements per posting (Bone et al., 2023). This shift is supported by analytic models relating wage or premium to skill supply, demand, and complementarity (Stephany et al., 2022). Green skills—environmental management, sustainable design, energy literacy—are also increasingly rewarded when combined with AI or technical competencies (Bone et al., 2023).

AI adoption induces positive spillovers, raising demand for complementary skills (analytical thinking, resilience, digital literacy) both within AI-specific roles and across adjacent occupations, with empirical studies demonstrating a “complementary effect” outpacing substitution effects (Mäkelä et al., 27 Dec 2024).

4. Core Green Skills and Interdisciplinary Demands

The evolving demand for AI and green skills encompasses:

• Technical proficiency in AI model design, including advanced efficiency methods (pruning, quantization, parameter-efficient tuning, green neural architecture search, lightweight modules) (Zhou et al., 2023, Verdecchia et al., 2023, Martínez-Fernández et al., 2023, Zhou et al., 2023).
• System- and hardware-level expertise for optimizing compute resource allocation (dynamic scheduling, elastic partitioning), server and storage management, and low-power inference (Zhou et al., 2023, Xie et al., 13 Jan 2025).
• Energy and carbon accounting, lifecycle assessment (LCA), product carbon footprint (PCF), and embedding sustainability as a core software/system requirement (Clemm et al., 1 May 2024, Cruz et al., 2 Jun 2025).
• Skills in federated learning and distributed systems that allow large-scale, energy-efficient AI (e.g., for IoT and edge devices, with quantization, sparsification, energy-aware client selection) (Thakur et al., 19 Sep 2024).
• Data management covering responsible sourcing, cleaning, labeling, and privacy-preserving data sharing frameworks, as well as synthetic data generation and blockchain-based provenance solutions (Xie et al., 13 Jan 2025).
• The ability to integrate environmental performance indicators into software, report emissions, and translate scientific advances into business and societal outcomes (e.g., green ESG metrics, digital transformation in emerging markets) (Cui, 3 Jul 2025).
• Policy and regulatory knowledge for incorporating environmental standards and reporting requirements into AI adoption at national and industry levels (Chowdhury et al., 20 Dec 2024).

Cross-sectoral and interdisciplinary skills are essential. Professionals must combine technical AI knowledge with systems engineering, environmental science, green energy management, and policy fluency to meet the twin transition of digitalization and decarbonization (Zhou et al., 2023, Cruz et al., 26 Jun 2024, Cruz et al., 2 Jun 2025). In practice, this also entails skills for smart urban planning, renewable energy optimization, and aligning data center operations with grid flexibility and renewable integration (Evans et al., 14 Mar 2025).

5. Educational and Institutional Responses

The imperative to embed sustainability in education and training is universally recognized (Verdecchia et al., 2023, Cruz et al., 2 Jun 2025). Curricula and professional development programs need to:

• Treat sustainability as a first-class objective, with intended learning outcomes that balance accuracy, performance, and environmental impact (Cruz et al., 2 Jun 2025).
• Offer project-based learning and open-source tools for energy assessment, benchmarking, and life cycle analysis of AI systems (Cruz et al., 2 Jun 2025, Verdecchia et al., 2023, Martínez-Fernández et al., 2023).
• Foster interdisciplinary and lifelong learning models that cover not only technical knowledge, but also environmental metrics, trade-off analysis, and the ethical/economic aspects of Green AI (Walsh, 2023, Cruz et al., 26 Jun 2024).
• Educate on the trade-offs between software implementation choices—such as programming language efficiency (with factors of up to 54× difference in energy use between C++ and interpreted languages), and how algorithm implementation can greatly impact real-world energy consumption (Marini et al., 31 Dec 2024).
• Prepare students and practitioners for emerging job roles that combine AI, green technologies, and complementary human attributes (such as teamwork, resilience, creativity), aligning with market trends where skill diversity is economically advantageous (Stephany et al., 2022, Mäkelä et al., 27 Dec 2024, Walsh, 2023).

6. Practical Implications and Policy Directions

Empirical evidence indicates that AI can reduce emissions and advance a green economy when integrated with effective safeguards and sustainable investments (Chowdhury et al., 20 Dec 2024). Distributed compute paradigms, such as spinning flexible AI/HPC workloads to follow renewable supply, can stabilize the grid and create new value streams, rewarding operators who possess joint AI and energy management expertise (Evans et al., 14 Mar 2025). At the same time, policy frameworks are required to incentivize green FDI, support smart cities, and catalyze upskilling in both AI and green domains (Chowdhury et al., 20 Dec 2024, Cui, 3 Jul 2025).

Corporate ESG performance is measurably enhanced by human-AI technology integration, facilitated by innovation in green technologies—an effect that is more pronounced in larger or state-owned enterprises and in developed regions (Cui, 3 Jul 2025). Addressing disparities in resources and capabilities, as well as tailoring educational and workforce development initiatives to regional contexts, is a necessary step for maximizing the benefits of AI-led sustainability (Cui, 3 Jul 2025).

7. Future Directions and Open Research Agenda

Ongoing research efforts highlight the need for:

• Standardized, reproducible, and accessible methodologies for energy and carbon assessment at varying system granularities (Cruz et al., 2 Jun 2025).
• Advancement of meta-methodologies and modular sustainability-aware architectures that accommodate hardware and workload heterogeneity (Cruz et al., 2 Jun 2025, Martínez-Fernández et al., 2023).
• Adoption of life-cycle-based system thinking frameworks that concurrently address model, data, server, and cloud impacts, leveraging AI for the automated optimization of its own sustainability profile—termed “AI4greenAI” (Clemm et al., 1 May 2024).
• Expansion of toolkits, benchmarks, and empirical studies for energy-aware AI, with rigorous field validation and industrial adoption (Verdecchia et al., 2023, Cruz et al., 2 Jun 2025).
• Broader and deeper interdisciplinary collaborations among software engineers, data scientists, environmental scientists, and policy stakeholders, ensuring that energy and resource optimization are embedded throughout the AI system lifecycle (Zhou et al., 2023, Cruz et al., 2 Jun 2025).

In summary, the demand for AI and green skills is intensifying due to environmental imperatives, technological complexity, labor market evolution, and the pressing need for sustainable digital transformation. The future workforce must be technically adept, systems-oriented, and environmentally literate, equipped to foster innovation that balances digital potential with ecological responsibility.