Making AI Sustainable: What We're Missing

Artificial intelligence has become one of the most transformative technologies of our time, but its environmental footprint remains largely unmeasured and poorly understood. As organizations worldwide race to implement AI systems across industries, a crucial conversation is emerging about the true cost of these technologies. Sasha Luccioni, a prominent AI researcher focused on environmental sustainability, argues that without significant changes to how we track and report emissions data, we cannot hope to create genuinely sustainable AI solutions that don't compromise our planet's future.

The problem begins with a fundamental gap in our understanding of how much energy AI technologies actually consume. Training large language models and other complex machine learning systems requires enormous computational resources, yet most organizations operating these systems rarely disclose their environmental impact. Luccioni points out that this lack of transparency makes it impossible for stakeholders—from regulators to consumers—to make informed decisions about which AI applications are worth their environmental cost. The research community has only recently begun systematizing the collection of emissions data, and even these early efforts reveal troubling patterns about the energy intensity of modern AI.

Beyond the technical challenges of measuring emissions lies another critical issue: understanding how people actually use AI in practice. Many organizations deploy AI models without clear insight into whether the technology genuinely improves outcomes or simply adds computational overhead to existing processes. This disconnect between deployment and actual utility means we cannot accurately assess whether an AI system's environmental cost is justified by its practical benefits. Luccioni emphasizes that creating sustainable AI requires understanding not just what energy is consumed, but whether that consumption delivers proportional value to users and society.

The infrastructure supporting modern AI systems presents another layer of complexity in the sustainability equation. Data centers that train and run sophisticated algorithms operate continuously, consuming significant amounts of electricity—often in regions where that power comes from fossil fuel sources. Even organizations that operate renewable energy facilities must contend with the reality that global energy grids remain partially dependent on non-renewable sources. Luccioni argues that we need better methodologies for calculating the actual carbon footprint of AI systems, taking into account not just direct energy consumption but also manufacturing impacts, cooling requirements, and the upstream emissions associated with building and maintaining the necessary hardware infrastructure.

Corporate accountability represents a crucial missing piece in the sustainability puzzle. Many companies deploying AI systems have made vague commitments to environmental responsibility but lack concrete metrics or reporting standards. Without standardized frameworks for measuring AI emissions, organizations face little pressure to optimize their systems for energy efficiency. Luccioni advocates for mandatory disclosure requirements similar to those that exist in other industries, ensuring that the environmental cost of AI development and deployment becomes as visible and scrutinized as other business impacts. Such transparency would create market incentives for companies to develop and prioritize more efficient AI algorithms and infrastructure.

The research community itself plays a vital role in advancing sustainable AI practices. Academic institutions and private research labs have begun publishing papers on the energy costs of training specific models, providing valuable benchmarks for the field. However, these efforts remain fragmented and incomplete, with many of the most resource-intensive projects conducted behind closed corporate doors. Luccioni emphasizes that open science and collaborative research are essential for understanding the true scope of AI's environmental impact. Without access to comprehensive data from major technology companies, researchers cannot build accurate models of how the field is developing or where the most significant efficiency gains could be realized.

The question of AI utility becomes increasingly important when viewed through an environmental lens. Some applications of artificial intelligence provide clear, measurable benefits that justify their computational requirements—such as using AI to optimize energy grids or design more efficient materials. Other uses of AI may be more marginal, adding minor conveniences without solving genuine problems. Luccioni argues that we need a cultural shift where organizations seriously question whether deploying an AI system makes sense given its environmental cost. This requires better tools for measuring AI usage patterns, understanding user adoption rates, and determining whether AI actually achieves its intended outcomes or simply creates the appearance of progress while consuming resources unnecessarily.

Policy frameworks will be essential for incentivizing change at scale. Governments around the world are beginning to develop regulations around AI safety and ethics, but AI sustainability remains an afterthought in most policy discussions. Luccioni advocates for policies that establish clear emissions reporting requirements for organizations deploying large-scale AI systems, similar to carbon accounting standards in other sectors. Such frameworks could drive innovation in efficiency, as companies compete to reduce their environmental footprint. Additionally, carbon pricing mechanisms that account for AI's emissions could help ensure that the technology is deployed only where its benefits genuinely outweigh its costs.

The path forward requires coordination across multiple stakeholders. Technology companies need to invest in more efficient algorithms and hardware, reducing the energy required for AI training and inference. Research institutions must establish standardized methodologies for measuring and reporting emissions, creating a common language for discussing AI's environmental impact. Policymakers should develop regulatory frameworks that make environmental accountability non-negotiable for organizations deploying AI at scale. Meanwhile, users and consumers should demand transparency about the environmental costs of the AI services they rely on, creating market pressure for sustainable alternatives.

Ultimately, Luccioni's message is clear: we cannot build genuinely sustainable AI without better data, better understanding, and better accountability. The technology industry has become accustomed to moving fast and asking questions later, but the environmental stakes of AI are too high for this approach to continue. By establishing comprehensive emissions tracking, understanding actual AI usage patterns, and creating policy frameworks that incentivize efficiency, we can begin to build an AI ecosystem that delivers genuine value without compromising our planet's future. The question is not whether sustainable AI is possible, but whether we have the collective will to make it happen.