By Amit Kapoor and Anandita Doda
AI as an economic story, not a tech story: where India is strong, where is it hollow
India has entered the AI era where the technology is moving from experimentation to everyday operations. A McKinsey survey finds that 88% of organisations used AI in at least one business function in 2025, yet only 31% were scaling it and just 7% had fully deployed it, suggesting an economy where AI is easy to try but hard to integrate. This gap between experimentation and integration is where the economic story begins for India.
Globally, AI is diffusing faster than ownership. While many economies are learning to deploy AI at scale, only a few nations, notably the United States and China, are accumulating the compute, capital, and research depth needed to capture most of the economic rents. These two remain the dominant AI superpowers, supported by general purpose models, advanced chips, and deep pools of private capital. The Stanford AI Index Report notes that US-based institutions produced 40 notable AI models in 2024 versus 15 from China, and that the US accounted for $109.1 billion in private AI investment in 2024 a scale advantage that compounds into talent concentration, compute access, and platform dominance. China continues to lead in several volume indicators such as AI publications and patents, reinforcing that the frontier is both technologically and economically concentrated.
Source: Stanford AI Index Report, 2025
India sits differently in this landscape as it has yet to build sovereign frontier models. The country is already a major AI economy by breadth and India’s share of AI publications in computer science reached 9.22% in 2023, comparable to the United States (9.20%). However, AI publications per capita in 2025 are far lower for India (4.45) than the US (22.90) and China (15.16), showing that India’s research intensity is diluted by population scale even when aggregate output looks large. The International Monetary Fund AI Preparedness Index places India at 0.492 (rank 72/174), below China (0.63) and far below the US (0.77), signalling gaps in digital infrastructure, innovation depth, and governance capacity that make scaling harder than piloting.
Source: OECD AI
India’s AI development lags not because adoption is absent, but because capability formation is constrained. India is not producing globally competitive frontier LLMs at scale and most general purpose LLM usage is through foreign platforms. Domestic efforts do exist, Krutrim, Sarvam AI and initiatives such as BharatGPT, but they are not yet the default model layer across large-scale Indian enterprise and public deployments. This is exactly why India’s scale advantage matters: if it can turn adoption into replicable platforms, it can capture productivity gains now while building domestic capability in the layers that compound.
India’s edge is not frontier training runs but it is implementation capacity through skills, integration talent, and the ability to diffuse workflow change across a very large economy. India is among the world’s largest producers of science and engineering graduates, giving it an unusually large pipeline for AI integration across firms and public systems. National Science Board (via NCSES) notes that India awarded 2.5 million first university degrees in science and engineering in 2020, the largest count globally. In 2023, 27.82% of India’s tertiary graduates were from STEM programmes. AI adoption requires redesigning workflows, training workers, building quality checks, integrating systems, and continuously improving outcomes and not just buying software. India’s large STEM pipeline will be a core advantage here.
That implementation capacity is already showing up on the ground in the form of frugal, problem-driven deployments across sectors. In health, non-invasive AI-enabled thermal imaging for early breast cancer screening in low-resource settings, and low-cost AI-assisted oral cancer screening devices deployed through primary health centres and outreach camps. In public goods and climate risk, it points to AI-based urban water management in Bengaluru and sensor networks plus machine learning for real-time landslide alerts in Himalayan regions. AI-enabled agricultural networks improving market access, price discovery and logistics for 1.8 million farmers across 12 states and a municipal education pilot in Pimpri-Chinchwad covering 18 classrooms across three schools, reporting improvements in engagement, teacher focus, and supervisory capacity.
These examples show what an Indian comparative advantage can look like: small, task-specific AI deployed in constrained environments where costs matter and multilingual access is essential, while domestic capability is progressively built in the sector-relevant layers. They also reveal a second economic truth: pilots are easier than platforms. If India wants diffusion to compound into national productivity gains, the enabling environment must reduce the cost of replication. That means treating AI less like software and more like infrastructure, while also being honest about market structure and capability gaps.
On the supply side, AI workloads are constrained by physical inputs like power, data centres, advanced chips, and GPUs. As per the World Bank, 73% of data centres are in high-income countries as of June 2025, China accounts for 11%, other upper-middle-income countries for 11%, and India for about 3%. At a global level, utilisation remains concentrated: high-income countries accounted for 58.4% of AI usage in April 2025, while upper-middle-income and lower-middle-income countries accounted for 22.5% and 18.7% respectively. The competitiveness opportunity for India lies in using AI widely but doing so in a way that is resilient to global concentration in hardware, capital, and proprietary systems.
Source: World Bank
The Economic Survey used an Agent-Based Model (ABM) of AI compute expansion to show what becomes binding as demand rises. As AI demand rises, GPUs, become the dominant bottleneck. India needs steady capacity building alongside compute-efficient design choices, which is why the India Semiconductor Mission and Semicon India programme aim to improve compute predictability through a ₹76,000 crore incentive framework with up to 50% fiscal support across the semiconductor stack.
After establishing compute as the visible foundation, data now becomes the invisible multiplier. Fragmentation in data availability and quality, lack of standardisation, and weak interoperability across systems and datasets prevents local ideas and innovation from adding up to real, country-wide strength. In economic terms, this behaves like a recurring transaction cost. Each institution rebuilds similar pipelines. Each successful pilot remains local. Productivity gains do not compound because systems cannot plug and play with each other. AI can be framed as a public good where the sovereign is a monetary stakeholder, akin in spirit to the rail-building mindset behind Aadhaar and UPI, so the state’s role is catalytic which is to coordinate, standardise, and unlock scale.
What separates isolated pilots from economy-wide productivity gains is infrastructure. Common data standards, interoperable systems, and reusable evaluation methods that make replication cheap and predictable. India’s comparative advantage in frugal deployment becomes dramatically more valuable when the plumbing cost of replicating solutions drops. A centralised code repository under the IndiaAI Mission, as proposed by the Economic survey is indeed the most practical idea. Government-hosted but community-curated, so that researchers, startups, and public agencies can reuse trusted components instead of rebuilding the same plumbing repeatedly.
An important point is that lagging sovereign AI development reflects not only shortages of compute or data, but structural constraints as well. India’s AI capability build-out is constrained by a weak private innovation engine and thin industrial depth: the private sector funds only about 36% of national R&D spending while government contributes roughly 64%, whereas in economies where R&D exceeds 2% of GDP, the private share is typically above 50%, a gap that matters because frontier AI progress depends on sustained, firm-led productization and scale-up. S&P Global Market Intelligence 451 Research estimates India’s data-centre IT load capacity at about 1.4 GW as of Q2 2025, with another 1.4 GW under construction; growing fast, but from a base that remains small relative to hyperscale AI trajectories. Over the next five years, more than 95% of India’s data centre capacity additions are expected to be driven by leased facilities, across both retail and wholesale segments, with the remaining share coming from hyperscale building dedicated AI infrastructure. On the ecosystem side, Tracxn reported AI startup funding falling by 53% from $305.9 million (FY 2023-24) to $143.6 million (FY 2024-25), which is exactly when global incumbents are scaling and market capture dynamics are intensifying.
Therefore, the bottom-up approach becomes a genuine advantage. Instead of making the national ecosystem hinge on a small number of compute-intensive frontier efforts, India can scale value through application-specific, small models that are computationally efficient, easier to fine-tune, and capable of running on locally available hardware such as smartphones and personal computers. This widens the set of actors who can innovate, including startups, research institutions, public agencies, and domain-specific firms, while reducing exposure to global GPU tightness or financial constraints.
At the same time, India can keep the long-term agenda of improving compute predictability and building interoperable data rails. In the short run, the economic play should be adoption-led: accelerate applied AI in priority sectors including education, help firms move from pilots to process redesign through shared playbooks and reference architectures, reduce trust and procurement friction through common evaluation norms, and expand affordable compute access for startups and smaller firms. Used this way, global models deliver productivity today, while domestic capability in fine-tuning, evaluation, security, and integration ensures India captures more of the value tomorrow.
The article was published with Open Magazine on February 20, 2026.
