Compute has become a matter of national priority. Without local capacity, countries risk permanent dependence on external providers, leaving them vulnerable to outages or sudden restrictions.
By Shruti Mittal
Shruti Mittal is a research analyst at Carnegie India. Her current research interests include artificial intelligence, semiconductors, compute, and data governance. She is also interested in studying the potential socio-economic value that open development and diffusion of technologies can create in the Global South.
Carnegie India’s Global Technology Summit Innovation Dialogue, co-hosted with the Ministry of External Affairs and an official pre-summit event to the India AI Impact Summit 2026, brought together several discussions on the future of AI development and innovation this year. Among its sessions was a panel discussion titled “The Spine that Powers Use-Cases: Compute and its Discontents,” which examined how infrastructure challenges shape the future of AI, and drew on experiences from building renewable-powered data centers in Africa, advising governments on digital transformation, and advancing AI and digital public goods in India. Three major themes stood out from the discussion.
Related Articles
Sovereignty and the Risk of Dependency
Compute has become a matter of national priority. Without local capacity, countries risk permanent dependence on external providers, leaving them vulnerable to outages or sudden restrictions. Developers in Africa have already seen their work disrupted when platforms went offline during national examinations. Startups face existential risks when global companies change rules or withdraw access.
For governments, the challenge is not whether to use cloud services, but how to balance them with sovereign infrastructure. Sensitive data such as defense, intelligence, or health records cannot be left entirely to external providers. At the same time, bursts of demand, as seen during the pandemic, may require cloud access. The task is to design layered strategies that combine sovereign cores, edge solutions, and selective reliance on cloud services.
Smart Engineering over Brute Force
Another theme was the need to use compute more intelligently. AI models can be resource-intensive, but not every problem requires maximum accuracy or constant generation of new outputs. In public health, for example, caching common queries or compressing models to run on mobile devices can reduce costs and extend reach.
Design choices matter. Chasing marginal gains in accuracy can exhaust budgets without improving outcomes. A screening tool that achieves 90 percent accuracy may be sufficient if it enables timely referrals. The broader lesson is that the focus must be on ascertaining the sufficient degree of compute required for arriving at a solution that can make meaningful impact. Compute is a means to an end and not the end in itself. This requires a shift in mindset: from building prototypes that consume vast resources to designing systems that can operate within the constraints of low-resource environments.
Models as Carriers of Culture
The panel emphasized that models are not neutral. They reflect the values and assumptions of the societies that build them. Respect for elders, for instance, is deeply embedded in African cultures but may not register in systems trained elsewhere. Over time, the dominance of external models risks flattening cultural diversity, embedding foreign priorities into everyday applications.
This raises a broader question: whose values will shape the digital future? If access compute remains concentrated in a few regions, the cultural imprint of those regions will dominate. Preserving diversity requires not only technical sovereignty but deliberate efforts to embed local languages, norms, and contexts into models. Otherwise, societies risk importing not just technology but cultural assumptions that may not align with their own priorities.
A Practical Agenda
The discussion at the GTS Innovation Dialogue offered a grounded view of the challenges ahead. Sovereign capacity, design, and cultural imprint are not abstract concerns. They are practical issues that determine whether AI solutions will remain stuck in pilots, or scale to meet real needs. For policymakers, the task is to invest in layered infrastructure that balances sovereignty with flexibility. For innovators, the challenge is to design with constraints in mind, ensuring that models remain viable in low-resource settings. And for societies, the imperative is to recognize that compute is not just about power, but about whose values are carried forward in the systems we build.
