ºÚÁϳԹÏÍø

Author: Eiji Matsuda

As the adoption of AI spreads, its potential to accelerate discovery and address global challenges is being tempered by mounting concerns over sustainability and equity. These questions took centre stage in early 2026, when The University of Tokyo and ºÚÁϳԹÏÍø co-hosted the seventh SDG Symposium on AI and Sustainability—Opportunities and Challenges for a Sustainable Future. 

Artificial intelligence (AI) is now deeply entrenched within research and is widely anticipated to accelerate scientific discovery and help tackle global challenges. Yet there is another side to AI that cannot be ignored—immense consumption of electricity and water, dependence on scarce materials, and the reproduction, or even amplification, of existing inequalities. 

At the SDG Symposium, leaders from academia, publishing, industry and international organisations looked at both AI’s costs and benefits from a variety of perspectives. They were joined by about 500 global participants.  

Emerging from the symposium was a clear recognition that â€˜sustainable AI’ is far more than a matter of technical and logistical optimisation; rather, it is a broader challenge that compels us to reconsider ethics, social structures, institutions, and patterns of resource distribution. 

In his opening remarks, Teruo Fujii, President of The University of Tokyo, stressed that this challenge demands collaboration among diverse actors, including leaders in the spheres of technology, industry, academia and policy. 

Physical constraints on the digital realm 

When interacting with AI through smartphones and laptops, it is easy to imagine that its computations occur somewhere nebulous in â€˜the cloud’. In reality, AI is profoundly physical. 

Magdalena Skipper, Editor-in-Chief of Nature and Chief Editorial Adviser for Nature Portfolio, presented stark data on the direct environmental burdens of AI’s rapid uptake. Searches using generative AI consume four to five times more energy than conventional web searches. Large data centres also require vast quantities of cooling water. In one reported case, a regional cluster of data centres consumed about 6% of the area’s monthly water supply. 

What may appear modest in global aggregate terms could pose a major threat to local communities. Skipper stressed the importance of making these hidden burdens visible, suggesting that just as light bulbs carry energy information such as wattage, AI chatbots might also display a comparable indicator. 

A further warning came from Tshilidzi Marwala, Rector of the United Nations University and an under-secretary-general of the United Nations. Manufacturing AI hardware requires rare earth elements, whose extraction imposes substantial environmental and social costs. Marwala challenged prevailing growth models that seek to maximise shareholder returns. He also cautioned that infinite growth cannot be achieved with finite resources. 

The recycling rate for rare earths is currently below 1%. Unless sustainability is embedded from the earliest stages of hardware design—as a third core metric alongside cost and performance—AI development may run up against the limits of the planetary environment. 

The semiconductor industry’s responsibility 

Positioned at the leading edge of the physical infrastructure underpinning AI, the semiconductor industry is not insulated from these concerns. Yuji Ogino, global head of sustainability at the semiconductor production equipment manufacturer Tokyo Electron, pointed to projections suggesting that surging demand for AI devices could drive the semiconductor market to US$1 trillion by 2030. 

AI’s progress has been propelled by the miniaturisation and increasing integration density of semiconductors, but this entails a steep rise in electricity consumption of data centres, along with a corresponding rise in carbon-dioxide emissions. Technological advances must come with a transition in data-centre energy sources from fossil fuels to low-carbon alternatives, Ogino argued. 

In his view, â€˜systemic governance’ requires simultaneous, integrated engagement with energy management and the environment, alongside ethics and human rights. In other words, governance cannot lag behind technological innovation. 

Elite capture or inclusive transformation? 

Another central theme? Fairness and equity, specifically, inclusion. Marwala noted that 80% of AI investment is concentrated in the United States and China. Left unchecked, such concentration risks expanding the gap between the advantaged and disadvantaged. 

Ayyoob Sharifi, Professor in sustainability at Hiroshima University, explored both the promise and the peril of AI in smart-city development. Sharifi urged society to move beyond a narrowly technical imagination. The essential question, he argued, is not simply whether something can improve efficiency, but whether it advances justice and sustainability. 

Interlocking SDG targets 

The 17 Sustainable Development Goals (SDGs) of the United Nations are interdependent. Xin Zhou, Director of the AI and New Frontier Group at the Institute for Global Environmental Strategies, highlighted the synergies and trade-offs involved. For example, progress in decarbonization (Goal 13) can yield spillover benefits for clean energy (Goal 7) and health (Goal 3); however, expansion of renewable energy may also compete with food systems or land use. 

Zhou expressed concern that policymaking too often remains siloed and argued that this is precisely where AI’s deeper value may lie. It could help visualise the complex interdependencies among water, energy and food systems, update risk maps in real time and support integrated policy decisions. 

But AI will not automatically repair fragmentation; it could just as easily deepen it. Whether AI accelerates sustainability or entrenches inequality and division, Zhou argued, will ultimately depend on governance. 

Physical AI and dialogue with society 

The symposium also considered the challenges posed by physical AI—AI moving beyond digital systems and into real-world space. Hironobu Takagi, Executive Director of the science museum Miraikan and senior researcher at IBM Research–Tokyo, highlighted the social dimensions of such a transition. 

Here, science communication is crucial. Social implementation depends on how sighted participants interact with the robot in demonstration trials and help co-create social rules needed for its use. 

From carbon reduction to social design 

The panel discussion broadened the debate. The moderator, Hiromi Yokoyama, Professor at the Kavli Institute of The University of Tokyo, reframed the scale of the problem. Sustainable AI, she argued, is not merely about reducing carbon footprints; more fundamentally, it concerns how societies design the social, economic and governance systems in which AI is embedded. 

In practice, the primary beneficiaries of AI often differ from those who bear its costs—whether in the form of depleted water and electricity supplies, or the broader social burdens of intensified surveillance. These asymmetries are unevenly distributed across both geography and generations. 

Regulation cannot rest with a single actor, the panel agreed. Rather, it requires a multilayered approach spanning the international community, nation states and industry. Yet it also acknowledged sobering realities: regulatory evasion through loopholes, and unavoidable disparities in AI performance for languages and populations that remain poorly represented in data. Future AI systems, the discussion suggested, should disclose such limitations as explicitly as the side effects of drugs are disclosed. 

A future built on inclusive participation 

Perhaps the clearest message was that AI development suffers from a profound lack of transparency, accountability and cross-border, multilevel cooperation. Marwala argued that the overwhelming underrepresentation of Global South languages and cultures in AI training data was an ethical failure. 

Addressing this challenge from the perspective of scholarly publishing, Antoine Bocquet, Managing Director of ºÚÁϳԹÏÍø Japan, pointed to another structural imbalance: the unequal distribution of knowledge that informs policymaking. , he noted, found that 78% of the research cited in policy documents originates from authors in the Global North, revealing a persistent geographical asymmetry in whose knowledge counts. 

Kensuke Fukushi, Director of the Institute for Future Initiatives at The University of Tokyo, closed the symposium by emphasizing the value of academic collaboration as the foundation for dialogue and shared rule-making in an increasingly fragmented world. In a divided world, he suggested, scholarly collaboration is essential. 

The road to sustainable AI will not be smooth. A system in which the benefits of AI are monopolized by the powerful and by a limited number of advanced, affluent countries, while environmental and social burdens are displaced onto future generations and vulnerable regions, demands correction. 

The central question is whether AI can be transformed from something that intensifies environmental strain and inequality into an infrastructure that supports a sustainable society. The answer lies not within algorithms alone, but in the wisdom and governance of the people who design, use, regulate and learn to live alongside them. 

Discover more from our SDG programme—featuring events, blogs and publications on sustainability and research.

Related content:

• How SDG research drives policy: A special report 
• Can open access help SDG research achieve greater impact? 
• Researchers embrace AI. The disclosure gap is the challenge 

¶Ù´Ç²Ô'³Ù miss the latest news and blogs, !