From Code to Copper: Why the AI Stack Must Begin with Electrical Infrastructure

For the past few years, the global AI discourse has been remarkably narrow.

AI stakeholders and commentators have talked about models, startups, automation, India's IT industry, job loss, AI chips, and Nvidia’s valuation. The conversation has oscillated between techno-euphoria and techno-anxiety. Artificial Intelligence has been framed as software — as code — as something that lives in servers and manifests as chatbots, copilots, and generative tools.

What has almost never been discussed is the electro-physical backbone that makes AI possible.

Yesterday, at the ongoing AI Impact Summit, the Adani Group unveiled one of the world’s largest integrated energy-and-compute investments, committing 100 billion dollars to develop renewable-energy-powered, hyperscale AI-ready data centres by 2035. By 2035, the investment is expected to catalyse an additional 150 billion dollars across server manufacturing, advanced electrical infrastructure, sovereign cloud platforms and allied industries, creating a projected 250 billion dollar AI infrastructure ecosystem over the decade, the Adani Group stated in an official release. The roadmap builds on AdaniConnex’s existing 2 GW national data centre footprint and expands toward a 5 GW deployment, creating what the Group describes as the world’s largest integrated data centre platform. The architecture combines renewable power generation, transmission infrastructure and hyperscale AI compute in a single coordinated ecosystem — with energy generation, grid resilience and high-density processing capacity to be developed in parallel.

This unprecedented-scale renewable-powered, AI-ready data centre expansion announced by the Adani Group opens up a dimension that has been largely absent from public debate. I do not see it merely as a capacity expansion announcement. I see it as a structural shift — and perhaps the first mainstream acknowledgement ever — that AI is not just a digital phenomenon. It is an electro-industrial system.

If India is serious about AI stack and AI sovereignty, we must not begin with algorithms, but with electrical infrastructure.

The Electrical Infrastructure Dimension of AI

Artificial Intelligence, at its core, is the large-scale conversion of energy into computation, and computation into intelligence.

Data centres are not abstract clouds. They are power-dense industrial facilities. GPUs are high-energy processors utilising megawatts of power to produce probabilistic outputs. Every training run, every inference call, every model update rests on high-capacity transformers, switchgear, transmission lines, cooling systems, storage systems, and grid stability.

The global AI build-out is currently constrained less by imagination than by infrastructure:

- Transformer shortages have lengthened project timelines.

- Grid interconnection delays have slowed data centre approvals.

- Cooling requirements have intensified as compute density rises.

- Renewable intermittency poses planning challenges for 24/7 workloads.

- Battery energy storage systems (BESS) have become critical to stabilizing supply.

In other words, AI capacity is bounded by electrical systems architecture.

When we discuss sovereignty in AI, we tend to think of domestically-developed LLMs and SLMs. But without secure energy generation and storage capacities, resilient transmission networks, high-capacity transformers, industrial cooling systems, and cyber-secure energy grids, "AI sovereignty" remains rhetorical.

The electrical layer is not auxiliary to AI. It is foundational.

The most underappreciated opportunity for India lies not only in hosting data centres, but in building the electro-industrial ecosystem that sustains them — including domestic manufacturing of transformers, power electronics, switchgear, thermal systems, and storage infrastructure. That ecosystem is durable, exportable, and transferable across sectors: from renewable grids to EV charging networks to semiconductor fabs.

To put it futuristically, AI demand can become the anchor tenant for a new phase of the electro-industrial revolution.

The Electrical Manpower Dimension: A Sociological Perspective

As a sociologist, I find the manpower dimension even more compelling.

Much of the AI conversation is preoccupied with white-collar automation. Will AI replace office jobs? Will coders be displaced? Will administrative roles shrink? These are valid concerns, but they overlook another transformation unfolding beneath the surface.

Electrical infrastructure is labour-intensive in its build-out phase and skill-intensive in its operational phase.

Large-scale AI infrastructure requires:

High-voltage electrical engineers

Grid systems specialists

BESS installation and maintenance technicians

Transformer manufacturing workers

Industrial cooling engineers

Transmission line crews

Hybrid AI–electrical integration professionals

If AI penetrates the industrial edge — into ports, factories, power plants, telecom networks, and logistics parks & corridors — we will need a new category of worker: the AI-electrical hybrid technician. Someone who understands control systems and machine learning. Someone who can maintain predictive maintenance systems on-site. Someone who can integrate inference clusters with physical equipment.

These are not gig-economy jobs. They are place-anchored, skill-intensive, socially-dignified roles.

Indian society has long struggled to bestow due dignity to the non-degreed technical vocations. AI-linked electro-industrial expansion offers a rare opportunity to revalorize advanced (non-degreed) technical workers. Apprenticeship programs, polytechnic modernization, AI-electrical curriculums, and field-based integration training could create a pipeline of high-quality, non-precarious employment.

If we are serious about inclusive AI growth, we cannot restrict our imagination to data scientists in megapolitan offices. We must include substation engineers, transformer fabricators, grid managers, and on-site AI integrators into the AI manpower picture. 

AI industrial sovereignty, to me, is as much about manpower architecture as it is about compute architecture.

A Focused Policy Framework

If this electrical dimension is to be sustained, it cannot rely on corporate initiative alone. A focused and disciplined policy framework is necessary.

First, India should incentivize domestic manufacturing of AI-grade electrical components — especially high-capacity transformers, advanced power electronics, industrial thermal systems, and BESS technologies — explicitly linked to AI infrastructure demand.

Second, renewable expansion planning must be aligned with compute corridor planning. AI infrastructure and generation capacity must be co-designed, not sequentially improvised.

Third, technical education must evolve. Dedicated AI-electrical hybrid tracks in polytechnic schools and engineering colleges can prepare a new generation of industrial AI professionals.

Fourth, standards for distributed industrial AI architecture should be developed. AI should not remain hyperscale-only. Industrial edge AI deployments need secure, interoperable frameworks.

Fifth, grid modernization must be prioritized. AI’s growth will increasingly stress transmission and distribution networks. Grid expansion planning must anticipate this convergence.

These steps don't necessarily require sweeping intervention, but more structural coherence.

The Demand Question: Will the Infrastructure Expansion Be Justified?

Infrastructure precedes demand. But infrastructure without sustained demand becomes stranded capital.

For AI demand to justify large-scale electro-industrial investment, it must expand along two dimensions: spatially and sociologically.

Spatial Expansion: The Industrial Edge

As I have argued in two previous blogposts, the next phase of AI expasion would be AI moving to the 'edge' — that is beyond centralized cloud dashboards to industrial equipments and components. Therefore:

- Factories must embed predictive maintenance systems. 

- Ports must deploy AI-assisted logistics coordination. 

- Power grids must adopt real-time balancing algorithms. 

- Warehouses must use intelligent robotics.

- Agricultural equipment must integrate sensor-driven optimization.

This shift to edge AI would create continuous inference demand — 24X7 operational intelligence rather than episodic model queries. It also necessitates distributed compute nodes, not just hyperscale concentration.

If India’s major industrial conglomerates adopt embedded AI across their assets, demand would become endogenous. It would arise from operational modernization, not digital experimentation alone.

Without edge penetration, hyperscale expansion risks overestimation.

Sociological Expansion: Consumer Absorption

AI must also expand sociologically. Productivity tools alone cannot sustain mass demand. AI must be embedded into daily life — into language, education, cultural media, identity creation, and local enterprise support.

India’s diversity offers both challenge and opportunity. Regional language AI, educational tutoring systems, agricultural advisory platforms, and culturally embedded digital assistants could generate durable demand, if social normalization and trust are achieved.

When technology becomes socially habitual rather than economically instrumental, demand stabilizes.

Without sociological absorption, AI risks becoming elite consumer infrastructure.

Risks and Uncertainties

Such massive (and unprecedented) capacity expansion, of course, carries multiple risks and uncertainties. 

Financing remains a primary risk. Large-scale AI infrastructure requires complex capital structures. Over-leverage, especially in a cyclical global environment, could create systemic vulnerability.

Overcapacity risk is real. If demand — industrial or consumer — does not scale as anticipated, data centres and associated electrical investments may underperform.

Chip and hardware supply remain volatile. Dependence on imported GPUs and advanced data-centre hardware exposes infrastructure timelines to geopolitical disruptions and trade policy volatility. 

Renewable energy intermittency also presents challenges. AI workloads demand energy reliability; balancing green energy commitments with stable compute requires careful integration of storage systems with energy grids.

Energy supply concentration could generate allocation politics. Prioritizing energy supply for AI compute over other sectors may create regulatory friction, if not transparently managed.

Finally, conglomerate concentration carries systemic risk. If AI-electrical infrastructure is dominated by a small number of large companies, economic power may centralize further, reshaping competition dynamics.

These risks do not invalidate the opportunity — but they demand sobriety.

Conclusion: Re-centering AI Around Electrical Infrastructure

We often speak of AI as if it floats in the cloud. It does not. It rests on high-capacity transformers, transmission corridors, power electronics, thermal systems, storage infrastructure, and grid resilience. It depends on engineers, technicians, grid managers, and industrial integrators.

If India seeks genuine AI sovereignty, we must look beyond models and applications. True AI sovereignty cannot not be secured merely by developing domestic LLMs or hosting hyperscale data centres. It can be secured when the electrical infrastructure and electro-industrial systems that sustain intelligence at scale are resilient, domestically-anchored, and technologically-advanced.

The real foundation of AI is not the algorithmic layer. It is the electro-industrial backbone that enables continuous computation — reliably, securely, and at scale.

Before there are models, there are transformers.

Before there is inference, there is grid stability.

Before there is intelligence, there is infrastructure.

The nations that understand this hierarchy — and build accordingly — will anchor, define, and lead the next industrial revolution.