Getting to the AI frontier: India must move from access to leadership

The AI Impact Summit demonstrated that India has made meaningful progress in building the foundations of an artificial intelligence ecosystem. Under the IndiaAI Mission, shared compute capacity has expanded, graphics processing unit (GPU) pricing has been lowered, teams are developing indigenous foundation models, and applications across agriculture, health, finance, and public services are being catalysed. This application-first approach has created momentum. The question now is whether this approach is sufficient for the strategic environment taking shape. 

The baseline of the AI economy is a G2 world dominated by the United States and China. Frontier model development requires tens of thousands of advanced GPUs, dense data centre clusters, reliable power, specialised talent, and multi-year capital commitments. Both the US and China are making annual investments in the hundreds of billions of dollars in frontier AI capabilities. Once compute, energy, and research ecosystems co-locate at this scale, advantage compounds. 

Leading AI researchers have warned that middle powers, such as India, cannot rely on diffusion of AI applications alone. The capital and talent required to train frontier systems are rising sharply, and access to the most advanced models cannot be assumed indefinitely. Even if applications spread widely, economic rents and strategic leverage accrue where frontier capability is built and continuously improved. The implication is not that every nation must match the United States or China, but that credible near-frontier capability is necessary to avoid structural dependency. 

Several middle powers have drawn this conclusion. France has backed Mistral as a frontier lab tied to sovereign compute initiatives. Japan has integrated AI into its economic security framework. The United Arab Emirates (UAE) has combined capital and energy advantage to build sovereign model programs and large-scale infrastructure. While none of these countries is seeking to match the US or China, they are building up their strategic leverage. 

India’s current approach is application-first and focused on democratising AI access. Companies such as Sarvam are building multilingual models suited to Indian conditions, and IndiaAI is expanding shared compute access. This is necessary but not sufficient. An application-first strategy has become a full-stack AI strategy for five reasons. 

First, value extraction is moving upward. Frontier model owners capture disproportionate rents through application programming interfaces (APIs), enterprise licensing, and ecosystem control. Building applications on externally controlled models risks long-term dependency. Second, affordability cannot be assumed. As AI becomes embedded in enterprise and public systems, pricing power may concentrate. Near-frontier domestic capability materially improves negotiating leverage. 

Third, national security considerations are inseparable from AI capability. Frontier models are dual-use systems relevant to cybersecurity, intelligence, logistics, and defence. Operational autonomy at the inference layer is useful, but strategic autonomy requires influence over training and updates. Fourth, long-term capability building requires depth. Frontier labs anchor research talent, generate spillovers into academia and industry, and create durable scientific capacity. Without concentrated nodes, ecosystems plateau at integration rather than advance core science. 

Fifth, training priorities reflect the interests of those who fund and control frontier labs. The most advanced multimodal systems are optimised for enterprise productivity, advertising, and global consumer markets. India’s needs differ. With hundreds of millions consuming entertainment in multiple languages, there is substantial demand for multimodal models that can generate and localise video, audio, and interactive content across Indian languages and cultural contexts. If training objectives are set elsewhere, optimisation will not prioritise Indian linguistic nuance or regional storytelling economics. Over time, this shapes which industries benefit most from AI advances. 

The obvious question is cost. A serious frontier program would likely require annual capital commitments in the low single-digit billions of dollars over five to seven years to fund large-scale GPU clusters, energy provisioning, and sustained research teams. In the context of India’s overall digital infrastructure and industrial investment plans, this is material but manageable. It represents strategic insurance rather than financial overreach. Moreover, we have many corporate groups that can build such a frontier capability working within a well-designed policy framework. 

This effort should be designed carefully. India does not need a proliferation of symbolic labs. We should strive to enable a small number of frontier-scale champions with ring-fenced compute access, milestone-based funding, and periodic performance review. Shared pools should continue to support ecosystem diffusion, but frontier training requires dedicated capacity and long-horizon planning. 

The strategy should also be embedded in a broader middle-power framework. In a G2 baseline world, pluralism will not emerge organically. India should work with France, the United Kingdom, Japan, Korea, the UAE, and others to build a distributed network of frontier capabilities that reach 80 to 90 per cent of global best performance on critical capability benchmarks. Such a network would reduce systemic dependence, create interoperability options, and influence governance norms more effectively than isolated efforts. 

India’s civilisational scale strengthens the case. With one-sixth of humanity and extraordinary linguistic diversity, India cannot remain only a user of AI systems shaped elsewhere. Artificial intelligence will increasingly mediate economic transactions, public services, cultural production, and national security. The AI Summit should therefore mark a transition from access to leadership. In the coming decade, countries that possess credible frontier capability will shape the terms on which AI is diffused. India must be among them.

 

---

The writer is president, Everstone Group, and visiting professor in practice at the London School of Economics. He is a former Union minister and Lok Sabha MP. Views are personal