The views expressed in this article belong to the authors and do not reflect the views of their employer or affiliated entities.

In February, New Delhi hosted the IndiaAI Impact Summit. Focusing on three pillars of people, planet, and progress, the summit brought together over 20 heads of state, 60 ministers and 500 global AI leaders. The summit produced several landmark outcomes: over 90 countries signed an AI Impact Summit Declaration, Sarvam AI launched two indigenous AI models, 13 private sector companies became parties to New Delhi Frontier AI Impact Commitments, and investment commitments topped USD 200 billion across infrastructure, foundational models, hardware and applications. Addressing the congregation in a plenary session, Prime Minister Narendra Modi outlined that India’s AI vision is clear — AI is a shared public good and India has what it takes to lead the charge on innovation and inclusion.

While the summit highlighted India’s ability to convene global stakeholders and articulate an ambitious AI agenda, it also provided a useful reference point for evaluating how these ambitions are being translated into practice. In this context, the IndiaAI mission, which is considered central to building India's AI future, provides an important lens for assessing progress. Available data indicate that less than 4% of India’s approved AI budget has been disbursed in the IndiaAI Mission’s first two years. This figure captures the central tension in India’s AI ambitions: what the country wants to build vs what it is actually building. But the difference between the grandeur of the summit and the pace of execution raises broader questions about what really is India’s AI ambition, and is it doing enough to meet it? The evidence, increasingly, suggests that the answer is no.

India’s grand AI ambitions and policy actions do not add up

The IndiaAI mission, the government of India’s flagship program to advance AI, focuses on an everything, everywhere, all at once approach: compute, innovation, datasets, application development, skilling, startups and responsible AI. The intent is comprehensive but appears to lack prioritization. Given India’s current capital intensity and institutional capacity, this distinction matters.

The US and China built dominant AI capabilities by prioritizing strength in certain layers of the AI stack at different stages, while progressively expanding toward broader stack integration rather than pursuing uniform, full-stack dominance from the outset. For example, the US anchored first in compute and private capital, China in state-directed deployment. India’s advantage lies in AI talent, application development and large-scale deployment. The issue is that these layers are treated as coequal rather than foundational. The risk is that India expends scarce policy attention and fiscal bandwidth thinly across the stack, without anchoring leadership in any single layer of the AI value chain, resulting in investment diffusion.

As a result, the fiscal picture is not ambiguous. In a written reply to the Rajya Sabha on the progress of the IndiaAI Mission, the Government disclosed that fund releases have lagged sharply behind approvals. Against a total approved outlay of over ₹10,300 crore over five years, only ₹21.79 crore was released in 2024–25 against revised estimates of ₹173 crore, and ₹379.15 crore in 2025–26 against revised estimates of ₹800 crore, with no funds released so far for 2026–27. As a result, less than ₹400 crore, under 4% of the total approved allocation, has been disbursed in the first two years of the Mission’s execution.

Moreover, the semiconductor program tells the same story. The recently released 24th Budgetary Report by the Parliamentary Committee on Communications and Information Technology (‘the Budgetary Report’) reveals significant under-utilization of public funds. Funding for semiconductor and display manufacturing incentive schemes was cut by 38.56%, declining from ₹7,000 Crore at the Budget Estimates stage to ₹4,300.38 Crore at the Revised Estimates stage. The Ministry itself recognized that the semiconductor manufacturing ecosystem is in a nascent stage, and companies are failing to meet policy conditions.

Such delays may also be attributed to the underdeveloped manufacturing ecosystem in the country. As Soumya Misra points out in The Print, while most states in India offer attractive incentives, they are only given for a limited period of time as opposed to the actual power and water supply requirements for a fab and Assembly, Testing, Manufacturing and Packaging (‘ATMP’) unit.

Private capital does not make up for this gap. As one takes a look at private funding of domestic AI capabilities, the numbers are dismal. Indian AI startups raised USD 1.34 billion in 2025 through 198 deals. This amounts to roughly only 0.6% of the total global funding pool of USD 225.8 billion floating around in the ecosystem. In addition to this, many startups are focused on creating AI applications rather than foundational AI models, raising investors’ concerns about long-term returns.

Recent macroeconomic research by Luisa Carpinelli, Filippo Natoli and Marco Taboga for the Centre for Economic Policy Research reinforces why such prioritization matters. Evidence from advanced AI economies shows that large‑scale AI investment does not translate into commensurate domestic economic returns when it is spread across layers without deep anchoring in any single segment of the value chain. A significant share of AI‑related capital expenditure leaks abroad through imports of hardware and infrastructure, while productivity gains accrue disproportionately to actors that control tightly integrated parts of the AI ecosystem, particularly compute and platform layers. Where investment outruns domestic value‑chain depth, returns appear muted in national accounts, even when headline spending is substantial.

In this context, attempting to advance all layers of the AI stack in parallel, without the capital intensity or institutional capacity to sustain them, risks producing precisely such leak‑prone outcomes. This concern is echoed in a 2024 Report by the Bank of International Settlements, which emphasizes that AI’s macroeconomic gains depend less on headline investment levels and more on coordination across data, compute, infrastructure, and institutions. Where investment is fragmented across silos or advances faster than complementary capacity, productivity gains remain narrow and volatile, despite rapid adoption.

Deeper tension between sovereignty, scale and market integration

Further, there is visible dissonance in the government’s approach towards AI advancement. For any country, and especially for India, a cloud-first approach is key to unlocking the power of AI. This is particularly relevant for harnessing AI use-cases to improve public service delivery — an outcome Nandan Nilekani has emphasized as the end-goal of AI Diffusion in India. However, current policy directions suggest otherwise.

On March 20th 2026, the MeitY released an office memorandum effectively directing all non-classified Government of India and state government data to be stored on government-owned or sovereign cloud infrastructure. Private providers, empaneled by the ministry, would only be allowed to store applications and data disclosable under the Right to Information Act, 2005. At the same time, substantial funds are flowing into India from major AI players to capitalize on market demand. The tension between a policy push toward a closed or sovereign cloud ecosystem and the promotion of large-scale investments, highlighted at the AI Summit, sends mixed signals about the country’s vision.

The deeper problem is the persistent conflation of data location and data control. Sovereign control is not automatically secured merely because data is physically stored within national borders. Data hosted in India but managed by foreign cloud providers may still be subject to foreign legal regimes, contractual asymmetries, or extraterritorial enforcement pressures. Conversely, cross‑border cloud arrangements, if paired with robust encryption, auditability, and enforceable contractual safeguards, may in some cases offer greater effective control than territorially rigid mandates. Treating localization as a proxy for sovereignty risks oversimplifying a complex technical and legal issue, while foreclosing more nuanced governance solutions.

What India must do

Excessive fragmentation imposes cumulative costs, eroding scale efficiencies, hampering interoperability, and raising compliance burdens for both providers and users. In an AI ecosystem where value accrues through network effects and shared infrastructure, the economic costs of isolation may compound faster than their security benefits. Policymaking, therefore, cannot rely on absolutist notions of either openness or closure. Instead, it must grapple with where fragmentation strengthens resilience, and where it simply imposes friction.

At present, the architecture has three compounding problems: investments are too thinly spread across all layers of the stack, execution machinery is not focusing on in-depth development and fund utilization and policy thinking on data governance appears against the cloud-first, open market approach that India’s own AI diffusion goals require.

India’s AI path requires focus on core fundamentals: democratizing compute capacity, deepening telecom infrastructure, funding R&D and adopting a data governance framework that will attract private capital that public capital cannot replace. Ultimately, for domestic AI capability to compound, these enabling conditions need to be built with the same urgency and rigor that characterized the organizing and execution of the country’s AI summit.