For three decades, India has been celebrated as an emerging technological power. It has one of the world’s largest pools of engineers, a globally respected information technology sector, a successful space program, and a fast-growing digital economy. Yet beneath these achievements sits a persistent contradiction: India trains enormous numbers of technically skilled people but has historically produced few globally dominant platforms, foundational technologies, or frontier research ecosystems of its own. In the age of artificial intelligence, where technological leadership increasingly determines economic, military, and geopolitical power, this contradiction has become harder to ignore, and, as the data below shows, it is now finally being addressed, late and unevenly, but in earnest.

The argument of this essay is that India’s innovation gap is best explained by institutional incentives (inadequate), capital structures (poor), and political economy (asymmetrical).

The Investment Gap, in Numbers

The single most-cited and most glaring problem is India’s chronically low R&D intensity. India’s R&D investment as a percentage of GDP stands at roughly 0.64%, far below China’s 2.41%, the United States’ 3.47%, and Israel’s 5.71%. South Korea spends close to 4.93% and Japan 3.30%. India is not merely behind the frontier economies, it also trails middle-income peers like Turkey and Egypt in R&D-to-GDP ratio despite ranking seventh in the world in absolute R&D spending, a sign of how much scale alone can mask intensity.

This gap is compounded by who is doing the spending. Private industry contributes only 36.4% of India’s gross R&D expenditure, compared with roughly 77% in China and 75% in the United States. In the countries that successfully climbed the technology ladder, the state seeded research and industry then took over financing and scaling it. In India, the government remains the primary funder decades after liberalization, indicating that the institutional conditions that make deep research commercially attractive remain underdeveloped. The issue is not simply a shortage of capital but an ecosystem problem: venture finance is concentrated in faster-return sectors, exit markets for research-intensive firms remain shallow, university-industry linkages are weak, and intellectual-property protection does not consistently translate innovation into durable economic rents. Together, these factors reduce the incentives for private capital to finance high-risk, long-gestation research at scale.

This pattern also reflects the incentives shaping Indian capitalism. A significant share of private investment has historically flowed toward sectors where profits derive from access to markets, land, regulation, natural resources, or financial intermediation rather than from sustained technological innovation. Under such conditions, firms rationally prioritize short- to medium-term returns over the uncertain payoffs of frontier research, producing a private sector that is often adept at commercialization and adaptation but less willing to bear the risks associated with original technological discovery.

The trajectory is improving. India’s Global Innovation Index rank rose from 81st in 2015 to 40th in 2023, and further to 38th in 2025, but the base it is improving from is exceptionally low, and the private-sector financing gap has not closed nearly as fast as the publication count has grown.

The Publication-Commercialization Gap, Measured

The thesis that India produces papers but not products is not just intuition, it shows up clearly in patent-to-publication ratios. India’s share of high-quality research articles climbed 44% between 2019 and 2023, reaching 1,494.27, yet in the same period China and the United States each produced roughly 23,000 and 20,000 high-quality articles, respectively, a gap far larger than the underlying difference in research spending alone would predict, suggesting structural barriers in how Indian research gets converted into protectable, monetizable output.

The country’s own flagship public research body illustrates the bottleneck concretely. The Council of Scientific and Industrial Research (CSIR), India’s largest network of national laboratories, filed about 250 Indian patents and 213 foreign patents in 2022-23, and holds a portfolio of 1,132 unique patents in force, of which only 140 have actually been commercialized. That is a commercialization rate of roughly 12% on its own active patent stock. This situation in CSIR’s own earlier internal reviews, has been treated as a long-standing institutional problem rather than a recent one. A government-commissioned review of CSIR’s technology transfer performance over 2007-2017 found the organization had filed 202 patents but published 659 papers in the period under review, roughly a 1-to-3 ratio of patents to papers from a body whose explicit mandate is industrial application. The same review noted CSIR has had to “form alliances with licensing firms abroad” simply to find buyers for IP generated with Indian public money, because no comparable domestic commercialization infrastructure existed. It also stated that foreign IP firms could reduce CSIR’s IP costs by at least 30% and enhance overseas patent filing capacity.

This is the empirical core of the “publish but don’t own” syndrome, and it is an institutional failure of technology-transfer infrastructure, evaluation metrics, and venture ecosystems, replicated across publicly funded research bodies in many developing countries. What is more striking, however, is that India has been slower than several of its peers to implement the corrective institutional reforms that elsewhere helped bridge the gap between scientific output and commercial innovation.

Countries, like South Korea, Israel or China, facing similar constraints invested in technology-transfer offices, strengthened university-industry linkages, aligned academic incentives with commercialization, and created financing mechanisms capable of carrying research from laboratory to market.

India’s challenge, therefore, is not simply one of scientific capacity but of institutional adaptation: the country has demonstrated an ability to produce knowledge, yet it has been less successful in constructing the organizational and financial architectures through which knowledge is transformed into intellectual property, productive firms, and technological power.

Brain Drain: Where the People Go, and Why

The flow of Indian scientific talent abroad is large, measurable, and has a clear economic logic behind it.

Over 1 million scientists and engineers of Indian origin now work in the United States, a number that has risen sharply over the last decade.
Nearly 1.6 lakh (160,000) Indian students are pursuing higher studies in the US alone, and over 1.3 million Indians renounced citizenship between 2015 and 2024, many in high-skilled sectors.
Crucially, at least one-third of IIT graduates, the products of India’s most selective and heavily subsidized engineering education, migrate abroad, according to NBER-cited research.
The pull factor is explicit in comparative survey data: in a 2019 international study on graduate retention, by the Center for Security and Emerging Technology, 89% of Indian-born AI PhDs in the US said they wanted to remain there despite long green-card waitlists, far higher than the 50% figure for Singapore-born PhDs, a gap researchers attribute to fewer scientific career opportunities in AI back home. Notably, Singapore-origin retention intent has been falling as Singapore’s own R&D investment in AI has risen (despite being a small island nation, almost matching India’s own investment in AI research), a natural experiment suggesting that domestic opportunity is the deciding variable.

Why would talent stay rather than return? India spends only about 0.65% of GDP on R&D against 2.5% in China, 4-5% in South Korea and 3.5% in the US, a gap that is itself cited as the primary driver of the talent outflow. India’s elite institutions also have a severe capacity bottleneck: in 2025 the top IITs offered only 17,760 seats against roughly 1.48 million JEE candidates, or about one seat for every 83 applicants, meaning even domestically, the system filters out far more talent than it can absorb, let alone employ at internationally competitive salaries afterward.

This is the brain drain story: a story of seat scarcity, pay differentials, weak domestic research infrastructure, and, as the next section shows, a slowly improving but still thin landing pad for returnees.

The “Virtual Brain Drain” – A Second, Less-Discussed Channel

There is a subtler version of the dependency problem that does not require anyone to emigrate. Scholars of science describe a “center-periphery” structure in global science, in which researchers in advanced economies disproportionately control research agendas, major funding streams, computing infrastructure, journals, and venture capital networks, while scientists in developing countries participate primarily as collaborators rather than agenda-setters. Some researchers term this “virtual brain drain”, where rich nations effectively use poorer nations’ research labor for projects whose intellectual ownership and commercial value accrue elsewhere, even when the researchers never leave home.

This is the international-collaboration dynamic that this essay, in particular, wish to highlight with respect to India, as it is real but under-acknowledged, and its mechanism is structural: who funds the grant, who owns the resulting IP under the collaboration agreement, whose journals confer the career-relevant prestige, and whose venture capital is available to commercialize the output. All of these depend on the contractual and institutional terms under which Indian science plugs into a global system whose financial and infrastructural center of gravity sits elsewhere.

The critical question, therefore, is whether Indian science policy recognizes this asymmetry as a structural feature of the global knowledge economy or continues to treat international collaboration as an unqualified good, without sufficient attention to how ownership, commercialization, and technological rents are distributed within it.

Where Sovereign AI Actually Stands – Update From the Last Twelve Months

Since early 2025, India has moved from near-zero domestic frontier AI capacity to a functioning, government-backed foundational model program:

India’s national compute capacity crossed 34,000 GPUs by April 2025, and later scaled to roughly 38,000 GPUs against an original target of just 10,000, a fourfold overshoot of the initial ambition.
Sarvam AI was selected in June 2025 to build India’s sovereign LLM ecosystem, an open-source 120-billion-parameter model, backed by 4,096 NVIDIA H100 GPUs secured through Yotta Data Services with roughly Rs. 99 crore in government GPU subsidies. Recently it raises $234 million funding at $1.5 billion valuation.
An IIT Bombay-led consortium, BharatGen, has received Rs. 900 crore under the IndiaAI Mission, the largest single allocation so far, and has released Param2, a 17-billion-parameter multilingual Mixture-of-Experts model aimed at Indic-language governance, education, and healthcare applications.
Of 506 proposals submitted to the IndiaAI Mission’s Foundational Model pillar, 43 specifically target large language models, and demand has scaled fast enough that what was a debate over needing 1,000 GPUs eighteen months earlier has become a landscape where at least 47 proposals each require more than 2,000 GPUs.
By March 2026, the program had expanded to a dozen participating organizations, including Tech Mahindra (whose in-house “Project Indus” Indic LLM predates the formal Mission), Fractal Analytics, and several university consortia.

This is a genuine, if recent and still fragile, course correction. It also illustrates the essay’s structural thesis precisely: progress arrived from a change in institutional design: direct state compute subsidies, ring-fenced funding pools, and explicit IP-retention requirements (“sovereign” was written into the program’s name deliberately, to ensure models are trained, deployed, and owned domestically rather than fine-tuned on top of foreign weights with foreign licensing terms attached).

The honest caveat: Despite some claims, recently, of the new AI models beings trained from the scratch, several of the models built under the Mission so far, including Sarvam-M and Fractal’s 14-billion-parameter model, are fine-tuned on top of foreign base models like Mistral and DeepSeek rather than trained from scratch. And Indian builders report that access to the most cutting-edge GPU clusters used by global frontier labs remains limited even as the overall GPU supply has improved. India has bought itself a seat at the table; it has not yet closed the compute or foundational-research gap with the US or China.

What the Evidence Supports

Pulling the threads together, a consistent picture emerges. The central constraints on India’s scientific and technological advancement are overwhelmingly institutional: insufficient investment in research, weak mechanisms for commercialization, incentives that encourage talent outflows, asymmetries within global knowledge networks, and the late but growing recognition of strategic technologies as objects of national policy. Considering the above discussions, the following five points are of particular importance:

India underinvests in R&D, at roughly a third of China’s intensity and an eighth of Israel’s, and the private sector carries less than half the funding burden it does in peer economies.
Public research institutions face a structural commercialization gap, evidenced concretely in CSIR’s own patent-to-paper ratios and its reliance on foreign licensing intermediaries.
Talented individuals rationally emigrate toward systems offering more research funding, faster career progression, and deeper capital markets, a pattern that tracks R&D intensity and one that shifts in real time as relative opportunity shifts (note Chinese overall retention rates in the US falling in the recent years as Chinese R&D investment rises).
International collaboration sometimes transfers de facto ownership of Indian-origin research output abroad through funding and IP structures: a “center-periphery” dynamic documented across the developing world, not unique to India. Yet that is hardly a reason for complacency. The more relevant question is whether India should accept this pattern as inevitable or emulate countries such as China, which builds indigenous capacity and engage globally without surrendering innovation ownership.
Sovereign AI is no longer a hypothetical concern. It is an active, well-funded national program with real compute, real models, and real institutional buy-in, even if it remains behind the frontier.

The countries that solved these problems – South Korea, Japan, China, Israel – did so through industrial policy, financing reform, and state-directed compute and capital. India’s own most recent improvement, the IndiaAI Mission, is following exactly that playbook: subsidized compute, ring-fenced capital, and IP-retention rules. That is the lever that has actually moved in the last eighteen months, and it is the lever future policy should keep pulling.

Conclusion

India’s technological challenge has never been a shortage of intelligence, talent, or ambition. It has been, and to a diminishing but still real extent remains, a problem of institutional alignment: too little capital chasing deep research, too few domestic pathways to convert publications into products, and a global science system whose centers of funding and prestige sit elsewhere. The last eighteen months show that when the state changes the underlying incentives, which are compute subsidies, ring-fenced funding, sovereignty requirements written directly into program design, the Indian institutions respond quickly and at scale. The ultimate test, therefore, is not scientific capability but institutional capacity: the ability to convert research excellence into enduring technological and economic power.