"India Lacks Frontier AI Capability; Compute Tax Proposed for Data Centres"
Frontier AI model
A large-scale AI system at the cutting edge of capability, requiring massive compute and data to train.
Scaling laws
Empirical rules showing AI model performance improves predictably with more compute and data.
India's IT and IT-enabled services sector contributes roughly 7–8% of GDP and generates over $315 billion in annual revenue. For decades, its competitive advantage rested on large pools of technically trained engineers delivering software development and back-office services at a fraction of Western costs. This billable-hours model is now under structural pressure. AI tools can automate tasks that previously required teams of junior engineers, and clients are demanding higher productivity at lower cost. Large IT firms are reporting growth of only 1–3%, while the pyramid staffing model that absorbed hundreds of thousands of engineering graduates annually is breaking down.
The strategic dimension of this challenge goes beyond the IT services sector. Frontier AI models, the most capable, large-scale systems trained on massive compute, are now being used in scientific research, cybersecurity, and defence. The global AI landscape is increasingly bifurcated. On one side, the United States is pursuing unilateral AI dominance, restricting exports of advanced models and chips to other countries. On the other side, multilateral forums such as the AI Action Summit held in Paris have tried to build inclusive AI governance frameworks, but major powers including the US declined to sign the declaration on inclusive and sustainable AI, signalling that geopolitical competition will shape AI access more than cooperative norms. This mirrors the structural challenge India faces in semiconductors: just as India depends on foreign chip fabrication for its India Semiconductor Mission goals, it depends on foreign compute hardware and frontier models for its AI ambitions. In both cases, the bottleneck is the same, physical hardware and the supply chains that produce it.
The India AI Mission was established to address this gap, providing a shared pool of 45,000 GPUs for Indian institutions and startups. However, this capacity is a fraction of what a single US frontier AI laboratory controls, illustrating that policy intent has not yet translated into strategic parity.
Frontier AI models are built on scaling laws: the empirical observation that a model's performance improves predictably with the amount of computing power and data used in training. The underlying algorithms are widely understood; what differentiates frontier labs from everyone else is raw compute scale. This means that building a competitive frontier model is primarily a resource problem, not a secrecy problem. A country that cannot assemble sufficient compute cannot produce a frontier model regardless of the quality of its researchers.
India's compute gap is severe. The India AI Mission allocated 4,096 GPUs to Sarvam AI, India's flagship model project. Frontier models are typically trained on compute clusters roughly fifty times larger. No algorithmic ingenuity can fully compensate for this gap. Meanwhile, large data centres are being built across Indian states, primarily to serve multinational corporations. These facilities carry significant computing capacity, but their physical location in India provides no automatic benefit to Indian institutions. …
Consider the following statements regarding the AI Action Summit held in Grand Palais, Paris in February 2025:
I.Co-chaired with India, the event builds on the advances made at the Bletchley Park Summit held in 2023 and the Seoul Summit held in 2024.
II.Along with other countries, US and UK also signed the declaration on inclusive and sustainable AI.
Which of the statements given above is/are correct?
India aims to become a semiconductor manufacturing hub. What are the challenges faced by the semiconductor industry in India? Mention the salient features of the India Semiconductor Mission.
Consider the following statements regarding the AI Action Summit held in Paris in February 2025:
1.The Summit was co-chaired with India and built on the Bletchley Park Summit (2023) and Seoul Summit (2024).
2.The United States and the United Kingdom signed the declaration on inclusive and sustainable AI at the Summit.
3.The 2026 AI Impact Summit in New Delhi was structured around seven Chakras (Pillars) including Access for Social Empowerment and Secure and Trusted AI.
Which of the statements given above is/are correct?
Match the following AI governance summits with their correct descriptions:
List I (Summit) — List II (Key Feature)
A.Bletchley Park Summit — 1. First in the AI safety summit series; focused on frontier AI risks
B.Seoul AI Summit — 2. Second in the series; advanced AI safety commitments
C.AI Action Summit, Paris — 3. Co-chaired with India; US and UK did not sign the inclusive AI declaration
D.AI Impact Summit, New Delhi — 4. Structured around seven Chakras including Secure and Trusted AI
Select the correct answer:
Consider the following statements regarding the India AI Mission:
1.It provides a shared pool of 45,000 GPUs for Indian institutions and startups.
2.Sarvam AI was allocated 4,096 GPUs to train India's flagship AI model.
3.The compute capacity of the India AI Mission's GPU pool is comparable to that of a single US frontier AI laboratory.
Which of the statements given above is/are correct?
Which of the following best describes 'scaling laws' in the context of frontier AI model development?
Consider the following statements regarding the proposed 'compute tax' for data centres in India:
1.It would require data centres to reserve a portion of computing capacity for a publicly administered national compute pool.
2.The reserved capacity would be physically relocated to government-owned facilities.
3.Its rationale draws on the logic of spectrum licensing, where private use of a strategic resource carries a public obligation.
Which of the statements given above is/are correct?
Want the full note, Prelims Facts & MCQ answers?
Create a free account — takes less than a minute.
Create free account