Designing capability: institutions and the future of emerging compute
Vilas Dhar
President, Patrick J. McGovern Foundation
Emerging compute is redefining scientific sovereignty, and societies now face a pivotal moment to embed equity and public purpose into AI’s foundations, says Vilas Dhar, President of the Patrick J. McGovern Foundation.
Published on December 16th, 2025
Societies are at a pivotal moment in shaping a human-centered, AI-enabled future, as rapidly evolving computational models begin to redefine how intelligence is created, governed, and accessed. These shifts, as illustrated in the Frontiers in Science lead article by Roy et al. (1) occur infrequently in the history of technology and early architectural choices made during these periods often persist, influencing who can participate in scientific, economic, and civic life. This is already evident in countries whose public institutions cannot replicate or evaluate AI systems deployed in health, education, or public administration because they lack the computational capacity required for independent assessment.
In my work with governments, multilateral bodies, and civil society organizations, I see a consistent desire to understand, evaluate, and shape AI systems in ways that reflect local priorities. Through efforts in global digital cooperation (2), scientific review (3), and national capacity building, I have also seen how access to compute and the capacity to govern it determine whether nations can help shape their technological and economic futures. This Policy Outlook argues that we are in a brief window to embed equity, public purpose, and shared capability before technical and geopolitical forces solidify.
Emerging compute as a determinant of public capability
Transitions in foundational technologies have long reshaped how societies distribute power, organize expertise, and allocate resources. Electrification, industrial research laboratories, and the early internet each transformed how knowledge was produced and shared. Today’s emerging computational infrastructure represent a shift of similar magnitude, influencing which institutions can build advanced intelligence systems and which nations can pursue innovation aligned with their development priorities.
National investments in advanced compute capacity, together with global efforts such as the United Nations’ Global Digital Compact (2), reflect recognition that computation now shapes scientific competitiveness and political autonomy. These disparities are especially visible in countries still developing their digital foundations, where gaps in infrastructure and trained personnel determine whether nations can contribute to AI development or depend on technologies created elsewhere (4, 5, 6). Through my work with multilateral bodies designing scientific review mechanisms and capability-building frameworks, it is clear that nations seek not only to deploy AI but also to evaluate, adapt, and govern it on their own terms (7, 8). Computation has become an emerging expression of scientific sovereignty. Recent national and regional initiatives, such as Europe’s EuroHPC Joint Undertaking and India’s AIRAWAT program, illustrate how investments in compute infrastructure directly shape a country’s ability to pursue frontier research, train domestic talent, and evaluate models developed elsewhere.
This capability-centered view positions compute as an institutional resource that shapes who can respond to societal challenges (9). Without careful design, it may consolidate in ways that limit which ideas and institutions can contribute.
A window of institutional opportunity
Periods of technological transition create openings in which norms and expectations can still be shaped. Before new computational architectures settle into proprietary platforms or strategic infrastructures, early decisions influence how capability is distributed, and which institutions are positioned to guide future developments.
Research on emerging computational infrastructure remains fluid, with no architecture holding global dominance, and no institution establishing binding norms that determine access or use (7). This creates a window of institutional opportunity in which societies can embed equity, transparency, and broad participation into the foundations of the computational era.
Institutions determine who can build, interpret, and govern intelligence by structuring access to scientific resources, guiding research priorities, and cultivating the skills needed to participate (3). In practical terms, these institutions include national research agencies that allocate compute resources, universities and public laboratories that host scientific infrastructure, and multilateral scientific bodies that set standards for evaluating and governing advanced AI systems. In periods of transition, these choices either widen participation in shaping the field or concentrate influence among a few.
Such windows do not remain open indefinitely. Once computational paradigms crystallize into commercial or geopolitical infrastructure, institutional inertia sets in and the ability to reorient diminishes sharply (10). If societies do not act during this period of openness, emerging compute will likely be shaped primarily by market concentration and strategic competition rather than commitments to justice or public purpose.
Global public goods and the architecture of shared capability
Shared responsibility in this context includes building scientific networks, creating regionally accessible compute resources, and developing governance arrangements that allow countries to contribute meaningfully to the evolution of these systems (5, 8). Regional compute initiatives such as EuroHPC demonstrate how coordinated investment can expand access to high-performance infrastructure for research communities that would otherwise be excluded from frontier experimentation. The capacity to build, adapt, or scrutinize advanced intelligence must extend beyond a narrow group of institutions if the next generation of AI is to reflect diverse perspectives and problem-solving traditions. Meaningful participation in this era depends on access to computational resources that support scientific inquiry, economic development, and civic reasoning (4).
Designing computation as a shared capability requires infrastructures where universities, nonprofits, and public agencies can experiment without prohibitive barriers. It also requires international cooperation so that countries still building their digital foundations can develop the expertise, institutional frameworks, and governance capacity needed to participate fully (6, 7, 9). In many national and regional capacity-building efforts I have supported, I see a common thread. Nations want the ability to build and evaluate models independently, to understand risks and opportunities in their own contexts, and to anchor technological decisions in their own social priorities.
Education and scientific training form an essential part of this architecture. This includes investing in programs that strengthen core engineering capabilities for early-career researchers to work with real compute infrastructure, and supporting interdisciplinary training that brings computer scientists into conversation with social scientists, ethicists, and public administrators, as well as cultivating analysts, policymakers, and community leaders who can interpret computational choices that shape public outcomes and who can guide institutions as they adapt to new forms of intelligence — a pattern seen in earlier global scientific efforts where coordinated investment created durable public goods (11).
Institutional forms are already beginning to take shape. The UN’s Independent International Scientific Panel on AI (10) will provide technical assessment and shared global insight into the capabilities and risks of emerging AI systems. In parallel, new financing mechanisms that bring together government and philanthropic support, such as Current AI, aim to expand global access to computing and scientific expertise (5). Yet these institutions will not be sufficient on their own. Creating forums where communities can imagine the institutions they need, and participate in shaping them, may offer a way to innovate social structures with the same ambition and urgency that drive advances in technology (8).
Conclusion
The developments highlighted in this issue show that the foundations of computation are being reimagined, creating a brief period in which the values and institutions guiding AI can still be shaped.
Meeting this moment requires institutions that understand how design choices influence who can participate in the future that follows, and leaders who recognize that technical advances carry social and political implications. Governments, research organizations, philanthropy, and civil society share responsibility for strengthening the public goods that allow communities to study, question, and contribute to advanced computational systems, as well as for building the educational and scientific pathways needed to govern them.
Acting during this window will require coordinated investment in regional compute infrastructure, transparent mechanisms for evaluating advanced models, and sustained support for the scientific and governance capacities that allow nations to operate and scrutinize emerging AI systems. The next computational era will not solely be defined by scientific progress, but also by the commitments societies make now.
This is a moment that invites collective engagement. The choices made during this window will influence the structure of capability for a generation and societies can guide the evolving foundations of computation toward a future that is more capable, more equitable, and more aligned with the people it serves.
Copyright statement
Copyright: © 2025 [author(s)]. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in Frontiers Policy Labs is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Generative AI statement
ChatGPT 5.0 was used for light editorial assistance during the preparation of this manuscript. All ideas, analysis, and substantive content are the author’s own.
References
Roy K, Kosta A, Sharma T, Negi S, Sharma D, Saxena U, et al. Breaking the memory wall: next-generation artificial intelligence hardware. Front Sci (2025) 3:1611658. doi: 10.3389/fsci.2025.1611658
United Nations. Global digital compact [online] (2024). Available at: https://www.un.org/techenvoy/global-digital-compact
United Nations. Governing AI for humanity: interim report. UN (2023). Available at: https://www.un.org/sites/un2.un.org/files/un_ai_advisory_body_governing_ai_for_humanity_interim_report.pdf
United Nations Development Programme. Digital strategy 2022–2025. UNDP (2022). Available at: https://digitalstrategy.undp.org/
Clark J, Marin G, Ardic Alper OP, Galicia Rabadan GA. Digital public infrastructure and development: a World Bank group approach/ Digital transformation white paper, volume 1. World Bank (2025). Available at: https://openknowledge.worldbank.org/server/api/core/bitstreams/bb08d389-cfbf-4417-9415-b9f65ef5c1f4/content
International Science Council. Preparing national research ecosystems for AI: strategies and progress. ISC (2025). Available at: https://council.science/publications/ai-science-systems/
Global Partnership on Artificial Intelligence. Responsible AI working group report. GPAI (2022). Available at: https://wp.oecd.ai/app/uploads/2025/05/gpai-responsible-ai-wg-report-2022.pdf
United Nations Educational, Scientific and Cultural Organization. Recommendation on the ethics of artificial intelligence. UNESCO (2021). Available at: https://unesdoc.unesco.org/ark:/48223/pf0000381137
Organisation for Economic Co-operation and Development. Recommendation of the council on artificial intelligence. OECD (2019). Available at: https://oecd.ai/en/assets/files/OECD-LEGAL-0449-en.pdf
United Nations. Terms of reference and modalities for the establishment and functioning of the Independent International Scientific Panel on Artificial Intelligence and the Global Dialogue on Artificial Intelligence Governance: Resolution adopted by the General Assembly on 26 August 2025 (A/RES/79/325). UN General Assembly (2025). Available at: https://docs.un.org/A/RES/79/325
International Human Genome Sequencing Consortium. Initial sequencing and analysis of the human genome. Nature (2001) 409, 860–921. doi: 10.1038/35057062
