India's Infrastructure Bottleneck Has Migrated

Piyush Goyal, India's commerce minister, was pitching the country's national grid to data-center investors last December, and the pitch was simple: nobody else has one.

Europe does not have a national grid. Even the US does not have a national grid. But India has a national grid. So we are a preferred destination for data centres. — Piyush Goyal

The numbers behind the boast are real. In the fiscal year that ended in March, India added 52,537 megawatts of generation capacity — a record — and the national power shortage fell to 0.03 percent, down from 4.2 percent a decade ago ^[1]. The country hit its Paris Agreement target of 50 percent non-fossil electricity capacity five years early, in December 2025 ^[2]. The national transmission grid now spans more than 5 lakh circuit kilometers ^[3]. On the physical layer — the wires, the plants, the electrons — India is closing the gap at speed. That is not the setup for a takedown. It is the setup for a migration. Consider solar manufacturing. India's module assembly capacity reached 210 gigawatts as of December 2025, a fourteen-fold increase since 2020 ^[4]. That is the number the government leads with, and it is real. But cell manufacturing capacity — the step upstream, where silicon wafers become the photovoltaic cells that modules are built around — stands at just 27 GW. The gap between the two numbers is filled by imports: 99 GW of modules and cells entered India in 2025 alone, with China serving as the dominant cell supplier ^[4]. In March 2026, Chinese solar exports hit a record 68 GW, and for the first time cell and wafer exports surpassed finished panel exports — a shift that reveals exactly what importing nations like India are doing: building domestic assembly lines while remaining dependent on Chinese upstream components ^[5]. India can put panels on rooftops at world-leading scale. It cannot yet make the cell that goes inside them. Reliance Industries paused domestic lithium-ion battery cell production after failing to secure equipment from China ^[6].

We are probably at a moment of peak Chinese dominance in the electrotech system, as the rest of the world starts to wake up and realise that this is the energy future. — Kingsmill Bond

In artificial intelligence, India leads the world in adoption: 62 percent of workers use generative AI tools, the highest rate globally, and the country tops the AI Advantage index at 53 points against a global average of 34 ^[7]. But only 30 percent of India's tech workforce possesses the skills to build AI systems, and just 15 percent of organizations are actually scaling AI ^[8]. IBM's India head Sandip Patel has said the country needs to raise AI workforce literacy from 30 percent to 60 percent to become what he called "the skill capital of the world" ^[8]. The people inside these industries already see what the headline numbers obscure. The American Chamber of Commerce in India identified the electricity grid as the single largest constraint on India's AI ambitions.

We urge energy policymakers to consider that the electricity grid is the single largest constraint on India's AI ambitions. — Ranjana Khanna

The constraint is not a shortage of megawatts in the aggregate. It is a quality problem layered on top of a quantity success. AI data centers do not simply need more power; they need high-reliability, grid-stable, concentrated power delivered to specific locations. Grid India's chairman has warned that data centers risk causing systemic grid disturbances through what he calls "silent exits" — abruptly isolating from the grid ^[9]. Cushman & Wakefield found that severe power shortages, land constraints, and tightening regulations are naturally restricting data-center oversupply, with grid power availability described as "the chief concern" ^[10]. The physical layer is built. The technological layer — the ability to deliver the right kind of power, to the right places, with the reliability AI workloads demand — is still under construction. Goyal's grid pitch worked. Microsoft committed $17.5 billion, Google $15 billion, and Amazon $7 billion in AI infrastructure investment, and the national grid Goyal was selling is the competitive advantage he framed to them ^[11]. But the commitments land on a system whose upstream dependencies trace the same shape as the ones that define its solar industry. India can deploy. It can assemble. It can consume. What it cannot yet do — across solar cells, battery chemistry, and AI model development — is produce the inputs that deployment runs on. The solar panel on the Indian rooftop, assembled in an Indian factory from a Chinese cell, by a workforce that uses AI tools it cannot yet build — that is where the constraint now lives.