Session
Subtheme
Organizer 1: Civil Society, Latin American and Caribbean Group (GRULAC)
Organizer 2: Civil Society, Latin American and Caribbean Group (GRULAC)
Organizer 3: Civil Society, Western European and Others Group (WEOG)
Organizer 2: Civil Society, Latin American and Caribbean Group (GRULAC)
Organizer 3: Civil Society, Western European and Others Group (WEOG)
Speaker 1: Fieseler Christian, Civil Society, Western European and Others Group (WEOG)
Speaker 2: Elena Estavillo Flores, Civil Society, Latin American and Caribbean Group (GRULAC)
Speaker 3: David Bedard, Government, Western European and Others Group (WEOG)
Speaker 2: Elena Estavillo Flores, Civil Society, Latin American and Caribbean Group (GRULAC)
Speaker 3: David Bedard, Government, Western European and Others Group (WEOG)
Format
Roundtable
Duration (minutes): 90
Format description: A 90-minute roundtable is the ideal format for this session, ensuring an interactive and solution-driven discussion on the geopolitical and technical challenges of AI infrastructure. Given the complexity of the compute divide, a traditional panel would limit engagement, whereas a roundtable fosters direct exchanges between policymakers, researchers, and industry leaders, allowing for collaborative problem-solving. The discussion will be structured around our policy questions, ensuring that all participants—both onsite and online—can contribute insights on governance models, Global South leadership opportunities, and lessons from our current geopolitical background. Our invited speakers will introduce key ideas in no more than 15 minutes, setting the stage for an interactive debate. The remaining time will focus on real-time exchanges, capturing diverse perspectives. This format ensures that feedback from the session directly informs the creation of a prototype for a Global Alliance for Computational Power, helping refine the initiative and expand stakeholder engagement.
Duration (minutes): 90
Format description: A 90-minute roundtable is the ideal format for this session, ensuring an interactive and solution-driven discussion on the geopolitical and technical challenges of AI infrastructure. Given the complexity of the compute divide, a traditional panel would limit engagement, whereas a roundtable fosters direct exchanges between policymakers, researchers, and industry leaders, allowing for collaborative problem-solving. The discussion will be structured around our policy questions, ensuring that all participants—both onsite and online—can contribute insights on governance models, Global South leadership opportunities, and lessons from our current geopolitical background. Our invited speakers will introduce key ideas in no more than 15 minutes, setting the stage for an interactive debate. The remaining time will focus on real-time exchanges, capturing diverse perspectives. This format ensures that feedback from the session directly informs the creation of a prototype for a Global Alliance for Computational Power, helping refine the initiative and expand stakeholder engagement.
Policy Question(s)
1. What are the key geopolitical and technical barriers preventing equitable access to computational power for AI development, and how can international cooperation help address them?
2. What lessons can be drawn from global public goods initiatives, such as GAVI, to design a multilateral framework that ensures fair distribution of compute resources needed for AI development?
3. How can Global South nations shape AI governance and infrastructure policies to reduce dependence on foreign compute providers and build sustainable local AI capacity? What lessons can be learned from the DeepSeek case?
What will participants gain from attending this session? Participants will gain a comprehensive understanding of the geopolitical and technical challenges shaping AI development, particularly the growing compute divide between the Global North and South. They will leave with insights into the limits of current approaches, such as national compute strategies and private-sector-led initiatives, and explore how international cooperation could provide more sustainable and equitable solutions.
The session will also introduce the concept of a Global Alliance for Computational Power (see white paper attached to this submission), drawing lessons from successful multilateral initiatives like GAVI. Attendees will engage with diverse perspectives from policymakers, researchers, and industry leaders, helping them understand how governance, funding, and policy mechanisms could be designed to democratize access to compute power. Beyond theoretical discussions, participants will walk away with practical policy recommendations and new frameworks for AI capacity building, ensuring that computational power becomes a tool for inclusive innovation rather than a barrier to progress.
SDGs
Description:
The ability to train cutting-edge AI models is increasingly concentrated in a handful of countries with massive computational power, creating a "compute divide" that limits AI innovation in the Global South. While AI development relies on algorithms and data, access to affordable compute infrastructure has emerged as the biggest bottleneck. Training a single large language model can cost upwards of USD 100 million, effectively excluding entire regions from shaping AI’s future. This will explore the geopolitical and technical challenges of compute access, drawing from recent developments such as the DeepSeek AI model and discussions at the AI Action Summit in Paris. It will assess the limits of existing models—such as national compute strategies and private-sector initiatives—and propose new multilateral approaches. A key focus will be the proposal for a Global Alliance for Computational Power, a cooperative framework inspired by GAVI’s success in expanding vaccine access. By pooling resources and fostering collaboration between governments, industry, and civil society, such an alliance could provide sustainable solutions for AI capacity building in the Global South. Participants will debate how this model could work in practice, addressing governance, funding, and policy mechanisms. Bringing together policymakers, researchers, and technology leaders, this session will challenge participants to rethink AI infrastructure as a global public good rather than a privilege of a few nations. It will also provide concrete policy recommendations for equitable AI development, ensuring that computational power is not a barrier to innovation but a tool for shared progress.
The ability to train cutting-edge AI models is increasingly concentrated in a handful of countries with massive computational power, creating a "compute divide" that limits AI innovation in the Global South. While AI development relies on algorithms and data, access to affordable compute infrastructure has emerged as the biggest bottleneck. Training a single large language model can cost upwards of USD 100 million, effectively excluding entire regions from shaping AI’s future. This will explore the geopolitical and technical challenges of compute access, drawing from recent developments such as the DeepSeek AI model and discussions at the AI Action Summit in Paris. It will assess the limits of existing models—such as national compute strategies and private-sector initiatives—and propose new multilateral approaches. A key focus will be the proposal for a Global Alliance for Computational Power, a cooperative framework inspired by GAVI’s success in expanding vaccine access. By pooling resources and fostering collaboration between governments, industry, and civil society, such an alliance could provide sustainable solutions for AI capacity building in the Global South. Participants will debate how this model could work in practice, addressing governance, funding, and policy mechanisms. Bringing together policymakers, researchers, and technology leaders, this session will challenge participants to rethink AI infrastructure as a global public good rather than a privilege of a few nations. It will also provide concrete policy recommendations for equitable AI development, ensuring that computational power is not a barrier to innovation but a tool for shared progress.
Expected Outcomes
This session aims to generate concrete insights on how to bridge the compute divide and design a sustainable, multilateral framework for equitable AI infrastructure. By engaging policymakers, researchers, and industry leaders, we will explore solutions to the geopolitical and technical barriers that limit access to computational power, drawing lessons from global initiatives like GAVI.
A key outcome will be the creation of a prototype for a Global Alliance for Computational Power. The session’s discussions will help refine this initial proposal, ensuring it responds to real-world challenges and stakeholder priorities. Additionally, it will serve as a platform to draw more actors into the initiative, fostering collaboration between governments, multilateral organizations, and the private sector. The insights generated will also inform future publications and follow-up events, strengthening global efforts to make computational power a shared resource rather than a competitive advantage.
Hybrid Format: With a track record of hosting three hybrid sessions at IGF since 2020, ITS Rio understands that an engaging discussion depends on seamless interaction between onsite and online participants. The moderator will alternate between onsite and online participants to encourage direct exchanges. A dedicated online moderator will monitor the chat, selecting and elevating remote interventions in real time, while the onsite moderator will actively incorporate those perspectives into the discussion.
To maximize interaction, our invited speakers will collectively speak for no more than 15 minutes, setting the stage for a dynamic roundtable discussion. The remaining time will be dedicated to participant engagement, using a guided discussion format where attendees respond to specific prompts tied to the policy questions. To ensure a lasting impact, we will use Padlet as a collaborative notepad, allowing all participants—especially those joining remotely—to contribute key takeaways and policy recommendations in real time.