Digital Futures Lab

The Global Dialogue should become a turning point in centering climate justice in the AI Governance discourse in multilateral forums. The development and deployment of AI technologies requires electricity, water, land, and critical minerals and elements for hardware components. Dependence on fossil fuels for running data centres is contributing to emissions, while the promise of renewable energy is pushing resource extraction in vulnerable Global Majority settings like Africa and Latin America. With increasing global investments in resource-intensive AI infrastructures for large language models and mining to feed technology and energy transition needs, marginalised and climate-vulnerable communities continue to bear the disproportionate costs of livelihood and land displacement, soil and water contamination, emissions and a rapid depletion of natural resources. AI innovation must lean towards low-carbon pathways and promote intergenerational equity, ensuring that infrastructure development of the present does not compromise resource availability and the quality of life for future generations. As a first step, companies must share data on the resource use of advanced AI models. States, industry and civil society must reach consensus on a mandatory disclosure framework to publicly report projected resource (renewable and non-renewable energy, potable water) use, and all scope emissions, with adequate traceability, accuracy and public accessibility across the AI supply chain. Parties must establish planetary redlines for AI development, promoting the development of low-resource technology like Frugal or Small AI with domain-specific uses. These redlines must be contextualised to the realities of the Global Majority with high fossil-fuel dependency, energy poverty, mineral extraction histories, and clean transition pathways that need sustained support in finance, technology transfer and dependable alternatives. The goal of establishing redlines must be to wean technology development away from its current high-resource trajectory, mitigating emissions and reducing dependencies on critical mineral and rare earth mining, potable water, land and labour.

• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

The four selected priorities are fundamental to ensuring the benefits of AI are evenly distributed, harms anticipated and managed, especially for Global Majority contexts. AI capacity building, including technical expertise and talent for R&D, capacity to interface meaningfully with AI systems and to understand its limitations, are important for equitable AI adoption. This includes the capacity and resources to address the socio-environmental costs of AI development across the supply chain and determine the trade-offs between efficiency and productivity gains over high emissions, natural resource use and aggravation of climate change. Bottom-up problem diagnosis, data collection and processing informed by ethical and legal considerations, product development in consultation with domain experts and end-users, processes for institutions to ask AI vendors the right questions and frameworks to monitor deployed systems are just some parts of the capacity building puzzle that are necessary pre-requisites to AI adoption. AI systems do not exist in isolation - humans interact and interface with technology in their own contexts of access and vulnerabilities. Digital divides, varying institutional governance frameworks, linguistic and cultural diversity are just some of the factors that determine how different stakeholders across the AI supply chain experience new technologies. Understanding these implications, such as discrimination due to non-representative data, is necessary to build AI systems that do not replicate patterns of exclusion and marginalisation. Environmental concerns must also be centred in such a whole-of-system approach to AI, interrogating the connected socio-environmental consequences of AI development. From a legal perspective, this must translate into the protection and promotion of human rights, including environmental rights, across the AI supply chain. The protection of these rights requires evidence, including transparency about the environmental costs of R&D - e.g., electricity required in one training cycle, water needed to cool down servers, are just some of the indicators that developers must report.

One significant cross-cutting issue not adequately addressed by the current themes is the governance of AI's socio-environmental impacts. While the Pact for the Future dedicates attention to environmental sustainability, the Dialogue's themes do not treat the intersectional socio-environmental lens as a distinct area of governance. This gap matters as effective AI governance must account not only for minimising environmental harm but for building resilience in the systems and communities most exposed to it. The absence of comprehensive data on AI's socio-environmental footprint across the full supply chain significantly impacts our understanding of the scale of AI's environmental footprint, from the mining and processing of critical minerals and rare earth elements, potentially under conditions of forced labour, soil and water contamination, health hazards and livelihood impacts, to the energy and water intensity of data centre operations, to end-of-lifecycle e-waste,. These impacts are disproportionately concentrated in Global Majority contexts, where not-in-my-backyard geopolitics and weak enforcement of legal protections leave communities with little recourse. Another emerging dimension of the climate-AI infrastructure nexus is the climate vulnerability of AI. Drone attacks on facilities in the United Arab Emirates illustrate how critical digital infrastructure can become a casualty of geopolitical instability. As companies urgently seek regional hubs outside the Middle East, countries such as India are attracting significant speculative investment in data centre development. Yet many of the Indian states emerging as data centre hubs are also among the country's most climate-vulnerable locations with resource scarcity. For developers, climate vulnerability will increasingly disincentivise investment or impose substantial costs in climate-proofing infrastructure, a tension that the Dialogue's current themes do not address. The Dialogue will need to treat this nexus of climate vulnerability, infrastructure resilience and socio-environmental accountability as a cross-cutting priority, not an afterthought.

A lack of socio-environmental oversight on AI infrastructure expansion such as data centres, undersea cable landing stations, and fibre optic networks risks creating infrastructure with permanent ecological consequences without clear restoration mechanisms, or benefit sharing for local communities. In resource-scarce contexts, this is particularly worrying, as finance and research are directed towards opportunities for global digital benefits at the cost of local resources and livelihoods. Some of these impacts can be foreseen given similar patterns in other geographies, while some of them are already being experienced in India. Data centres are consuming fossil-fuel generated electricity in a complex ecosystem defined by uneven distribution and transmission networks and complicated electricity generation regulation. India is a fossil-fuel dependent economy. Reports from Mumbai show the government's plans to extend the life of coal plants to meet data centre needs. This will slow down progress on reducing greenhouse gas emissions. Furthermore, data centres use diesel generators for power backup, which are one of the biggest emitters in the electricity sector. Many hyperscale data centres are located in cities such as Mumbai, Bangalore and Chennai experiencing severe water shortages and compromised groundwater levels. Communities from states like Andhra Pradesh are resisting land acquisition for data centre projects, concerned about water use, electricity consumption and negative health impacts. Nickel mining in countries like Indonesia, supporting green-tech and AI pipelines, has caused water-source contamination, degraded forests and disrupted the lives of indigenous communities. India does not have a definitive policy on management of e-waste produced by data centres. India is one of the world's largest producers of e-waste - a majority of this waste remains untreated and is mostly handled by workers in the informal economy, posing health hazards and labour law violations.

The Dialogue can facilitate deliberations and work towards consensus building across several interconnected fronts in AI and environmental governance. Stakeholders must agree that the current negative costs of AI development do not disproportionately burden indigenous communities in the Global Majority, whose land and natural resources form the foundations of AI infrastructure and reduce the livability of the planet for future generations. An intergenerational equity lens in AI governance is key to equitable and sustainable futures. - As a foundational principle, AI innovation must pursue low-carbon pathways - adopting the least resource-intensive and emission-generating methods for the development and deployment of AI. This could mean promoting innovation in Small or Frugal AI, with clear uses and outcomes. - Environmental impacts must be centred (as opposed to current marginalisation) within mainstream conversations on AI safety and ethics, requiring sustained consensus-building across governments, industry, and civil society. - Mandatory minimum disclosure requirements for companies must be identified and enforced, ensuring that environmental costs are transparent and comparable across actors. While we acknowledge that information alone is not sufficient to address AI's resource consumption problem, it is a necessary step in understanding its magnitude, examining specific points of high-resource use and counter any false solution claims surrounding AI's climate potential. - Verifiable restoration mechanisms must be established to recover degraded lands post resource extraction, supported by finance and enforceable guidelines. This must occur alongside explicit prohibitions against greenwashing, so that mitigation commitments translate into measurable ecological outcomes rather than reputational cover. - Circular economy principles must govern e-waste management across the AI hardware lifecycle, prioritising repair, reuse, and responsible disposal over planned obsolescence.

The Global Dialogue can engage with emerging initiatives such as the Just Energy Transitions Network based in India, which brings together experts from disciplines in energy, climate policy and technology to discuss and deliberate on the linkages between digital transformation and clean energy transitions. Associations like the Friends of the Congo and the SIRGE Coalition are actively working towards protecting the rights of indigenous communities and natural resources in contexts of just transitions and AI development. Resources such as the Code Green media series - (https://www.digitalfutureslab.in/projects/code-green), developed by Digital Futures Lab, offer context-specific knowledge at the intersection of AI and environmental justice in the Global South, with a focus on Asia. They provide insights into research emerging from climate-vulnerable communities and centre justice and equity concerns in the AI-climate discourse. As a global forum, the Dialogue can create channels for evidence-sharing, connecting community-driven efforts to standards-setting and regulatory discussions, and supporting the development of local capacity for independent assessment of socio-environmental impacts of AI.

A variety of stakeholders are well-placed to contribute to the AI and environmental givernance discourse: - governments can establish policies for data centres that align with the current and estimated realities of their natural resources, - industry bodies can develop green infrastructure mechanisms, civil society and science and technology researchers can document the impacts of infrastructure on local communities, - academic and technical communities can develop socio-environmental impact assessment frameworks. To ensure diverse perspectives at the table, the Global Dialogue can set up focused thematic tracks on the socio-environmental implications across the AI supply chain, moving beyond electricity and water to land, livelihoods, labour and health implications. Tracks should include representation from policy, technical, legal and ethical perspectives. The following approaches can be used to enable meaningful cross-cultural participation at and around the Dialogue. - Multilingual written submissions. Participants should be able to contribute written submissions in their preferred language, with translation support provided for wider dissemination. This removes a significant barrier to participation for contributors from non-English-speaking contexts and ensures that a broader range of knowledge and experience can enter the deliberative process. - Pre-Dialogue synthesis and circulation. Key recommendations on environmental governance emerging from written submissions and oral inputs gathered during pre-Dialogue consultations should be synthesised and shared with all participants in advance of the Dialogue itself. This would allow participants to arrive informed by the full breadth of perspectives gathered, rather than encountering them for the first time during sessions. - Roundtable and fishbowl discussion formats. Plenary-style formats tend to concentrate voice and visibility among a narrow set of participants. Roundtable and fishbowl structures offer a more equitable alternative, enabling moderation responsibilities to be distributed across geographies, languages and professional backgrounds. When designed with care, these formats create the conditions for diverse knowledge systems to actively shape the conversation, rather than being nominally represented while remaining structurally marginal.

Global discussions on AI governance continue to underrepresent actors who are closest to how AI systems are experienced in practice. This includes - Civil society and grassroots organisations. Given the nascency of critical data centre studies and opacity around development, communities are under-prepared to ask questions about resource use and negotiate for their benefit. However, the proximity of civil society organisations (e.g., climate activists, land and livelihoods researchers) to affected communities position them to identify risks in real time that are not visible in technical or policy forums. Environmental defenders are also one of the highest persecuted groups in the world, with threats to their life and curtailment of their freedom of movement. This can make their participation in international forums physically difficult. - Youth representatives need to be actively involved in discussions about their future, based on principles of intergenerational equity. The finite nature of natural resources and irreversible nature of damages from AI infrastructure necessitate the presence of future generations in discussions that will impact them. To ensure inclusion of stakeholders, the following mechanisms are proposed: - Travel support and documentation assistance. Financial barriers and administrative obstacles, including the cost and complexity of obtaining travel documents, disproportionately exclude participants from Global Majority contexts. Dedicated support for travel costs and visa and documentation processes would meaningfully reduce these barriers, enabling a broader and more representative range of voices to attend in person. - Regional and in-country events. For participants unable to travel to Global North host countries, whether due to financial, logistical or legal constraints, organising Dialogue sessions and related events in Global South countries would ensure that environmental defenders and community activists are not effectively excluded from the process. These events should be designed as substantive deliberative spaces in their own right, with outputs feeding directly into the Dialogue's formal proceedings. - Live-streaming with privacy safeguards. Broadcasting deliberations in real time would extend access to participants and observers unable to attend in person, significantly expanding the geographic and demographic reach of the Dialogue. However, live-streaming must be implemented with robust safeguards to protect the privacy and freedom of expression of attendees, particularly environmental defenders and activists operating in contexts where public visibility may carry personal risk. Clear protocols governing consent, anonymisation and data handling should be established prior to any live broadcast.