India has been building the foundations for participation in frontier AI. The IndiaAI Mission had expanded shared compute capacity to more than 45,000 GPUs by June 2026, while the Government of India selected 20 indigenous foundation-model proposals for support under the Mission, comprising 12 large multimodal models and eight small language models. India also has a substantial base of AI researchers and inventors. Yet the first wave of globally notable AI models has been concentrated elsewhere. Stanford’s 2026 AI Index records 59 notable models from the United States and 35 from China in 2025, with industry accounting for more than 90 percent of notable models globally. India does not appear among the leading producers at comparable scale.
The larger policy question extends beyond generative AI. How can India build technological capability while an emerging technology is still taking shape, before its dominant applications, architectures and leaders become established? Once a technological trajectory begins to settle, leading firms accumulate research depth, specialised infrastructure, skilled teams, capital and market relationships that reinforce their position. India therefore needs institutions capable of learning and scaling while the frontier is still open, without depending on an ability to predict which technologies will ultimately prevail.
From Inputs to Capability
India has a substantial base of technical talent and a growing role for industry in research and development (R&D). Stanford identified about 50,460 AI authors and inventors associated with India in 2025, second only to the United States among the countries it tracked. The private sector accounted for 51.8 percent of India’s total R&D expenditure in 2023–24. India’s latest reported gross expenditure on R&D, for 2023–24, stood at 0.84 percent of GDP, compared with China’s 2.80 percent in 2025. Although the reference years differ, the gap in overall research intensity remains substantial.
The challenge is not only one of resources. India’s R&D ecosystem also faces delays in funding, rigid financial rules, procurement constraints, limited researcher mobility and weak pathways from research to technology transfer and validation. In frontier technologies, these frictions can slow the movement from an idea to experimentation and from experimentation to real-world testing.
India has also begun assembling new instruments, including the ₹1 lakh crore Research, Development and Innovation Scheme, expanded IndiaAI compute access and mission-oriented programmes under the Anusandhan National Research Foundation. The policy challenge is to make these investments work together as a system that can sustain experimentation, connect research with application and build capability over time.
From Participation to Technological Agency
Frontier technologies develop through repeated experimentation. Teams test competing approaches, learn from failure and redirect resources as evidence accumulates. This process builds more than individual technologies: it creates specialised teams, research programmes, infrastructure, industrial relationships and institutional know-how.
This is the basis for technological agency: the capacity to understand critical technologies deeply, adapt them to Indian conditions, originate important components and retain meaningful choices among competing technological ecosystems. It allows India to decide where indigenous capability is essential, where adaptation or partnership is sufficient, and where resources should shift as evidence changes.
Foundation models illustrate the principle. Capability in Indian-language models, data, evaluation, adaptation and deployment can strengthen India’s ability to shape how these technologies are used instead of simply adopting systems developed elsewhere. The aim is to build sufficient capability in strategically consequential parts of the technological stack while remaining open to adaptation and partnership elsewhere.
Building an Innovation System That Learns
India is increasingly assembling the instruments needed for frontier innovation. The Research, Development and Innovation Scheme provides long-term financing for high-risk, high-impact technologies; I-STEM is expanding access to publicly funded research infrastructure; and the Technology Development Board supports the development and commercialisation of indigenous technologies.
These instruments need to function as parts of a learning system. Funding should be linked to stage-appropriate technical milestones and periodic review, allowing promising projects to attract greater support while weaker approaches are redirected or stopped. At the same time, technical failure that generates useful learning should not be treated as programme failure. The purpose of review is to reallocate resources as evidence improves, not to discourage experimentation.
Infrastructure access should be predictable, and greater mobility across universities, public laboratories and start-ups can allow expertise to move towards promising problems. For frontier technologies, the speed with which researchers can access facilities, assemble teams and test ideas can influence whether capability develops early or only after technological pathways have hardened elsewhere.
The same principle should extend to adoption. For technologies with potential public applications, government can define difficult problems and clear performance requirements, allow competing teams to demonstrate solutions and create pathways to field validation. Challenge-based procurement can give promising technologies a demanding first user while generating evidence about their performance. Adoption then becomes part of the learning process rather than merely its final stage.
Measurement should reflect this shift. Funds sanctioned, equipment purchased and laboratories established capture inputs; stronger indicators show whether those inputs are becoming capability. These could include the time from funding to experimentation, progression from prototype to validation, technologies licensed or deployed, private capital mobilised after public support, and whether projects are redirected when technical evidence changes.
Testing the System Before the Frontier Settles
Physical AI, AI for science and quantum technology offer useful tests because their applications, architectures and industrial structures are still evolving. Each presents an opportunity to see whether India can assemble the teams, infrastructure, validation pathways and early markets needed before technological trajectories become established.
The National Quantum Mission, with an outlay of ₹6,003.65 crore over eight years, provides one such test. Its significance will ultimately depend on the durable capabilities that emerge from public investment: specialised firms, intellectual property, skilled teams, experimental infrastructure and early users.
The broader lesson applies across emerging technologies. India does not need to predict every winning technological trajectory in advance. It needs an innovation system capable of recognising promising ones early, learning from evidence and concentrating resources before technological leadership becomes difficult to dislodge.