India’s Hard Choice: Strategic Autonomy or Dependence?

Representative image

• India risks strategic issues from reliance on energy, chips, and AI infrastructure.
• Building domestic capacity in energy, semiconductors, and AI is key for autonomy
• Decisions in the next 5-7 years will shape India's global power and independence

Did our AI summary help?

The EU-India trade agreement is welcome progress. Delays around a U.S. trade deal need not cause anxiety if patience yields better outcomes. India must approach global partnerships with confidence and long-term resolve. But our success will depend increasingly on internal strength: building leadership across the AI stack, reforming institutions, reducing friction, improving execution, and becoming globally competitive in innovation and scale.

Global partnerships follow proven capability, not sentiment. That capability must be built at home.

The New Logic of Power

What looks like geopolitical disorder (fractured diplomacy, weaponised trade, automated warfare) is not chaos. It is the exhaustion of an older world order. Artificial intelligence didn’t entirely invent this shift, but it has accelerated it so decisively that familiar frameworks no longer suffice.

For decades after World War II, power was distributed across markets, institutions, and networks no single actor controlled. Economic growth came through trade integration. Supply chains optimised for cross-border efficiency. That world is receding.

Three shocks exposed the fragility: the 2008 crisis revealed interdependence as contagion, COVID-19 shattered efficiency-optimised supply chains, and systematic weaponisation of energy proved that resource dependence equals strategic vulnerability.

But these were symptoms of something deeper. AI crystallised a fundamental transformation: it made integration of foundational capabilities more valuable than access to any single one, and drastically compressed the timeframes in which advantages are gained or lost.

Power now reorganises around a different principle: whoever controls the most complete stack of interdependent capabilities increasingly sets the terms for everyone else.

The Six-Layer Power Stack

Power today reorganises around six interdependent layers:

Energy: Not just consumption, but capacity to generate persistent surplus at lowest cost. A single large language model training run consumes as much electricity as 1,000 Indian homes use annually. Nations without energy surplus either cannot build AI capacity at scale, or build it at economically unviable costs.

Physical and digital infrastructure: The logistics networks, fiber optics, electricity grids, and transportation systems that determine whether resources move at the speed algorithms demand. Infrastructure used to be about long-term development. Now it is about real-time responsiveness.

Compute capacity: Data centers and processing power to train and run AI systems. Global AI compute capacity is doubling every seven months. Countries without sovereign compute infrastructure create dependencies in everything AI touches: finance, logistics, medical diagnosis, military operations.

Semiconductors: The ability to produce, or secure guaranteed access to, advanced chips. Taiwan’s TSMC alone produces over 90% of the world’s most advanced processors. Nvidia has nearly monopolised high-end GPUs used for training and running artificial intelligence models. This concentration makes chip access a chokepoint for everything above it.

Data and algorithmic systems: Not data volume, as many in India believe, but frameworks and technical capacity to capture value from it. The question is whether data trains domestic models and builds domestic capability, or flows outward to train systems owned elsewhere.

Military power: Increasingly automated and AI-enabled. Precision targeting requires compute, semiconductors, energy, data, and infrastructure working in integration. Deterrence is no longer just about force size. It is about speed of intelligence processing and execution.

Here is what makes this a stack: each layer depends on the ones below it, and weakness in any layer constrains everything above. You cannot build AI capability without energy and chips. You cannot modernise your military without all five layers functioning reliably. You cannot secure any single layer if others are controlled by potential adversaries.

In the distributed world, you could be strong in some areas and weak in others, trade filled the gaps. In a stack-centric world, gaps compound into structural dependence.

Why Everything Feels Unsettled

The transition feels chaotic because different parts move at different speeds. Algorithms evolve in months. Compute scales in years. Chips take half a decade. Energy systems take even longer.

This explains what confuses traditional analysis: Energy is strategic again because it underwrites everything. Semiconductors are contested because they bottleneck compute, AI, and modern weapons. Geography matters because physical infrastructure cannot be virtualised. Diplomacy feels transactional because relationships reorganise around “Can you disrupt my stack?” not “Do we share a set of values (say democracy or participation in United Nations bodies)?”

Major powers are responding accordingly. The United States is consolidating a tightly integrated stack: domestic energy, semiconductor control through the CHIPS Act, compute concentration in American hyperscalers, platform dominance in AI models, advanced military-AI integration. China is assembling a counter-stack: manufacturing scale, the world’s largest renewable build-out, state coordination, hardened military capabilities integrated with domestic AI.

Between them lie structurally incomplete economies: Europe has industrial sophistication but lacks energy security and leading-edge chips. Japan and South Korea have chip capability but limited energy and market scale. Gulf states have energy and cash surplus but lack the technology stack above it.

India’s Position: Consequential but Incomplete

We enter this transition with genuine strengths: 1.4 billion people, the world’s fourth-largest economy, strategic Indo-Pacific geography, political continuity, and over 1.5 million engineers produced annually.

But our constraints cannot be wished away.

Energy: We import over 85% of our crude oil and about half of our natural gas. Coal provides 70% of our electricity. While renewable capacity expands rapidly, demand grows faster than clean supply can scale. Without surplus, cheap, reliable energy, we cannot build compute infrastructure at competitive cost.

Semiconductors: We have strong chip design capability but zero cutting-edge manufacturing. Government approved $10 billion in semiconductor subsidies, but even optimistic projections put our first meaningful fab online around 2027-2028, producing chips already two generations behind the frontier.

Compute: We have growing data center capacity, but sovereign AI training infrastructure remains limited. Most Indian AI development happens on cloud platforms owned by American or Chinese companies. We have no indigenous large language models competitive with GPT-4, Claude or even DeepSeek. We build applications on foundational models trained elsewhere, using compute infrastructure controlled elsewhere.

Military AI integration: We possess nuclear deterrence and conventional force capability, but AI integration into force structure remains a work in progress. Autonomous systems and networked warfare capabilities are under development, but not yet at the integration level visible in U.S. or Chinese programs.

From a market-size worldview, we appear well-positioned. From a stack-centric worldview, we are not yet structurally secure. Both are true. Confusing one for the other would be costly.

The Real Risk: Comfortable Dependence

Our greatest risk is not exclusion from global systems. It is integration without reciprocity.

Consider a future where data generated here trains models we license back as users, where talent builds systems for companies headquartered elsewhere, where platforms run on compute infrastructure beyond our control. This would not look like failure. It would look like participation.

But dependence accumulates through rational choices that each make sense individually but compound into structural constraints visible only when conditions change. The test is not ideological. It is operational: Can critical systems function if cooperation becomes conditional?

Strategic autonomy has never meant isolation. South Korea imports virtually all its energy but built semiconductor capability that gives it leverage oil reserves never could. Israel relies on American military systems but developed indigenous intelligence and cyber capabilities ensuring independent action when necessary.

What We Must Do

We do not need to dominate the AI-defined world. We must ensure we never become structurally dependent across all stack layers simultaneously, especially those determining military credibility, economic sovereignty, and the ability to make independent decisions in crisis.

Energy policy is national security: Build capacity to generate reliable surplus domestically through nuclear, solar, hydrogen, and emerging technologies, reducing import dependence below 50% by 2040.

Compute is infrastructure: Build sovereign AI training capacity sufficient to ensure critical applications, defense, finance, healthcare, governance, run on domestically controlled systems.

Data governance translates into retained value: Data generated in India should preferentially train models that build Indian capability through reciprocal agreements where data access trades for technology transfer, not just service fees.

Semiconductor strategy is realistic but relentless: We may not build 2-nanometer fabs soon. But we can build secure domestic capacity in legacy chips (28nm and above) covering 70% of applications, while securing guaranteed access to leading-edge chips through diversified partnerships with Taiwan, South Korea, Japan, and emerging U.S. production.

Military modernisation is inseparable from AI integration: Buying fighter jets maintains operational capability. Building domestic AI-defense integration sustains capability across decades.

Institutions are designed for speed: Regulatory sandboxes, fast-track procurement, and empowered implementation agencies are requirements now. This shift in mindset will have to travel across all our institutions.

Every major policy decision should pass a simple test: Does this reduce dependence in a critical stack layer? Build compounding capability? Create reversible dependence or lock in structural subordination?

The Choice Ahead

The AI age is not just a technological revolution. It is a reorganisation of power: more material, more integrated, more rapid, and more unforgiving than the distributed world it is replacing.

We still have room to maneuver. Our market size, strategic location, and technological talent provide leverage. But these can be depreciating assets if not converted into stack capability. China made the most of the past three decades. We don’t have the luxury to ruminate over missed opportunities. The next five to seven years may define the next half century.

The comfortable path is to optimise for today. The strategic path is harder: accept near-term costs, build foundations that will not pay off immediately, resist dependencies that look efficient but create structural vulnerability.

The decisions being made in 2026 about energy investments, semiconductor push, data governance, defense modernisation, and institutional reform will determine whether in 2047 we are a great power that controls our destiny or a large economy operating within constraints set by others.

The world is not necessarily losing order. It is learning a whole new grammar of power. India’s future will be decided by how clearly and how seriously we choose to speak it.