Economic Survey Admits Frontier AI Race is Too Costly for India

India has effectively conceded that building frontier AI models on the scale of global leaders like OpenAI, Google and Anthropic is out of reach, warning that trying to do so would drain public finances with little payoff. 

In a blunt statement, the government, in the Economic Survey 2025-26 report, says, “Attempting to close this gap would involve prohibitive fiscal costs towards what is increasingly becoming an unsustainable approach to AI development.”

It is the clearest signal yet that India does not plan to chase the race for massive foundational models that define the current AI frontier. 

Instead, the Survey argues that the frontier is already locked up by a handful of firms with capital, compute, data, energy access, and political leverage to secure large infrastructure and specialised hardware.

The government also flags a low-probability but high-impact risk, explicitly comparing it to 2008. It warns of a 10%-20% chance of a cascading shock where financial stress from leveraged AI infrastructure bets collides with technological concentration and geopolitical disruption. 

According to the Survey, recent AI investments are built on optimistic timelines, narrow customer bases, and long-duration capital commitments, making them vulnerable to a sharp correction. Such a correction would not halt AI adoption, but it could tighten liquidity, trigger risk aversion, and spill into broader capital markets. 

It further stressed that if financial stress were to coincide with trade disruptions or geopolitical escalation, capital flows could weaken abruptly, pushing economies into defensive postures. While the probability is described as residual, the warning is blunt. “The macroeconomic consequences could be worse than those of the 2008 global financial crisis.”

The Survey notes that over the past year, policymakers have gained clarity on how uneven the AI landscape has become. While AI tools are spreading across companies and sectors, the ability to train advanced foundational models is “highly concentrated,” favouring firms that can marshal vast resources. 

Export controls on advanced processors only deepen this divide, making frontier model development even harder for countries like India.

The Union Government frames the challenge as structural rather than temporary. “Most countries may end up participating in AI primarily as users, while a few will shape the technology's trajectory, standards, cultural leanings, and pricing,” it says. 

For India, the trade-off is stark. Scarce public capital can be spent chasing frontier-scale models with uncertain returns, or it can be deployed toward domain-specific systems aligned with domestic priorities in health, agriculture, finance, governance, and industry.

The Survey links this strategic choice to broader economic risks. Control over data, compute, models, and standards is increasingly concentrated, raising concerns about technological dependence and the future of India’s IT services sector. 

Firms that once relied on India’s labour advantage may need fewer people as AI matures, risking a hollowing out of the country’s traditional value proposition if adaptation lags.

On jobs, the Survey strikes a more cautious tone. Citing studies from Yale and European researchers, it says there is no evidence yet of large-scale AI-driven job losses. Labour is being complemented in the near term as firms integrate AI into workflows. Still, it warns that productivity gains from simple augmentation have limits. 

Over time, the labour intensity of growth may fall, especially in white-collar services that sit closest to AI adoption.

Constraints on compute receive particular emphasis. AI data centres demand enormous amounts of electricity and water, placing stress on grids and public resources. The Survey quotes US energy expert Mark P. Mills to underline the scale of the problem, noting that the energy used to launch a rocket is now consumed every day by a single AI-infused data centre. 

For India, where power, finance, and water remain binding constraints, indiscriminate scaling of compute comes with high opportunity costs.

Taken together, these arguments form the backbone of a clear policy pivot. Rather than pursuing frontier model supremacy, the Survey makes the case for a bottom-up AI strategy focused on smaller, application-specific models that run on limited hardware and decentralised compute. 

Such an approach, it argues, lowers entry barriers, reduces dependence on foreign systems, and spreads value creation across startups, research institutions, public agencies, and domestic firms.