Quick answer
Economic Survey 2025-26 identifies six AI trade-offs; 73% of data centres by count are in high-income countries, underscoring unequal infrastructure.
AI ecosystem trade-offs are choices between competing uses of scarce resources and competing policy objectives. The Survey identifies six linked tensions, arguing that India should widen useful adoption while avoiding dependence, exclusion and indiscriminate frontier-scale expenditure.
Concept diagram
Every AI strategy makes choices
Three balances illustrate strategic choices; their tilt does not measure a preferred allocation.
- Frontier capability and practical diffusion compete for scarce resources.
- Openness, scale and inclusion involve benefits as well as costs.
Basics: where value and control sit
Foundation model versus application
- A foundation model is trained on broad data and can support multiple tasks. An application combines models with domain data, interfaces and workflows to solve a particular problem.
- An application developer may buy inference instead of undertaking large-scale training. This lowers entry costs but may retain dependence on the model provider.
Fixed costs and lock-in
- Training infrastructure, specialised GPUs and compliance systems impose substantial upfront costs. Large incumbents can spread fixed costs over more customers.
- Vendor lock-in arises when switching providers becomes costly because data, interfaces or workflows are not portable.
Data chart
Chart XIV.2: Data-centre distribution by count, June 2025
- High-income countries account for 73% of centres by count.
- Centre counts do not measure their GPU capability, size or ownership.
Source: Economic Survey 2025-26, Chart XIV.2 (World Bank) · Selected categories printed in the Survey chart total 98%; no unreported residual is assigned. Shares refer to counts, not compute capacity.
Show the data (Per cent)
| Period / item | Share |
|---|---|
| High-income countries | 73 |
| China | 11 |
| Other upper-middle-income | 11 |
| India | 3 |
Six trade-offs in the Survey’s diagnosis
| Tension | Risk at one extreme | Balanced response |
|---|---|---|
| Frontier vs application | Scarce resources chase frontier scale | Prioritise domestic domain problems |
| Scale vs inclusion | Productivity outpaces worker reabsorption | Sequence adoption with augmentation |
| Open vs proprietary | Lock-in or fragmented quality assurance | Openness with stewardship |
| Compute vs resources | Power, water and finance displaced | Distributed efficiency and selective scale |
| Regulation vs innovation | Fixed compliance costs stifle entrants | Clarity and proportionate obligations |
| Autonomy vs integration | Dependence or costly isolation | Resilience within global networks |
Translate trade-offs into project appraisal
Test social returns
Compare productivity benefits with resource opportunity costs and distributional effects.
Protect contestability
Use interoperable systems and switching options rather than permanent provider dependence.
Calibrate safeguards
Match obligations to risk while allowing smaller innovators to experiment.
Common exam traps
Statements examiners like to twist — learn the correct version.
Open-weight means fully open-source.
Releasing weights does not necessarily release training data, code or unrestricted usage rights.
Strategic autonomy requires complete technological isolation.
The Survey explicitly recognises that complete self-sufficiency is neither feasible nor efficient.
Prelims focus
- Recall all six trade-offs and distinguish open-weight from proprietary models.
- Chart XIV.2 measures data-centre counts as of June 2025.
Mains angle
- Evaluate the distributional consequences of rapid AI diffusion in a labour-abundant economy.
- Explain how interoperability can support both competition and strategic autonomy.
Flip & recall
Answer in your head first, then tap the card to check.
Practice MCQs
Original Prelims-style questions with explanations.
0 / 3 correct · 3 left
Q1.Chart XIV.2 reports concentration by:
Q2.Which best reflects the Survey’s strategic autonomy approach?
Q3.Why can common compliance rules have unequal effects?
