AI SUPERPOWER IS FINE....BUT INFRA COSTS ?

 

BIZ MASALA

INDIA WANTS TO BE AN AI SUPERPOWER

WHO IS GOING TO PAY THE INFRASTRUCTURE BILL?

THE FORENSIC INFRASTRUCTURE EDITION

Hyderabad data-centre infrastructure. Real-world buildings—not abstract AI—are the starting point of this story.

Photo supplied for this Biz Masala edition; facility identified by the editor as CtrlS Hyderabad data-centre infrastructure. CtrlS describes its Hyderabad portfolio as AI-ready and lists DC3 at 13 MW.

AI does not run on algorithms alone. It runs on chips, racks, power, cooling, networks, buildings—and capital.

By Murthy ASR

 

THE FORENSIC QUESTION

India's AI story is often told as a story of models, startups and talent. But the harder question sits underneath: who pays for the physical machine that allows those models to exist at scale?

The answer is not one cheque and not one ministry. India's emerging AI infrastructure is being assembled through several overlapping pools of money: public AI funding, semiconductor incentives, data-centre investment, cloud and hyperscaler commitments, power infrastructure, and private capital. These pools should not be added together as if they were a single 'AI bill'. They finance different layers of the stack.

₹10,371.92 crore IndiaAI Mission outlay over five years

45,000+ IndiaAI shared GPUs reported by June 2026

₹1,27,500 crore Semicon 2.0 budget outlay

57 TWh Deloitte estimate for India's data-centre electricity demand by FY2030

Biz Masala thesis: India can become an AI superpower only if it simultaneously becomes an infrastructure superpower—or finds a sustainable way to rent that infrastructure from the world.

 

1  |  THE FIRST BILL: COMPUTE

Conceptual visual: the AI factory. The image is illustrative, not a photograph of a specific Indian facility.

The first physical bottleneck is compute. IndiaAI's shared compute pool has expanded rapidly: government material reports more than 38,000 GPUs onboarded in March 2026 and 45,000+ GPUs by June. That is a major increase in accessible national compute, but it is not the same thing as saying India owns 45,000 cutting-edge AI accelerators.

The distinction matters. The IndiaAI number is a shared compute ecosystem figure, while frontier AI companies, hyperscalers and private AI factories may operate separate hardware. The strategic issue is therefore not simply 'How many GPUs does India have?' but 'How much high-end compute can India reliably access, where is it located, and who controls it?'

Original Biz Masala analysis. IndiaAI shared-compute figures; not a census of every GPU in India.

The rack is becoming the factory

A new class of AI infrastructure is appearing around high-density GPU clusters. Recent industry announcements point to deployments measured not merely in servers but in thousands of accelerators and tens or hundreds of megawatts of supporting infrastructure.

The unit of AI infrastructure is shifting from the server to the power-and-cooling envelope around the server.

 

2  |  THE BUILDINGS BEHIND THE AI BOOM

A physical data-centre building: the unglamorous shell around the AI economy. Use only with verified facility attribution.

India's data-centre market is entering a scale-up phase. Deloitte's 2026 assessment puts installed capacity at about 1.5 GW in 2025 and estimates roughly 10 GW by 2030. CBRE reporting in October 2026 suggests India could approach 2 GW by the end of 2026.

Those figures cover the wider data-centre industry—not AI alone. That distinction is essential. Cloud, enterprise workloads, digital services, data localisation and AI all compete for the same physical infrastructure.

Original Biz Masala analysis using Deloitte's 2025 base and 2030 estimate, plus CBRE's 2026 market estimate. These are industry estimates, not AI-only capacity.

Hyderabad is a useful case study

CtrlS Hyderabad data-centre infrastructure. CtrlS lists Hyderabad DC1 at 6 MW, DC2 at 7.4 MW and DC3 at 13 MW, and describes its Hyderabad facilities as AI-ready.

Hyderabad illustrates why the infrastructure story is broader than an AI lab. CtrlS says its Hyderabad facilities are engineered for AI workloads, including liquid-cooling readiness, water recycling and redundant connectivity. The city is also seeing a much larger pipeline: current market tracking indicates substantial capacity under construction and planned.

 

3  |  THE BILL YOU CANNOT SEE: ELECTRICITY

Every AI data centre is ultimately a power project. GPUs consume electricity, electricity becomes heat, heat has to be removed, and the system must continue operating through peaks, outages and maintenance.

Original Biz Masala analysis. Deloitte estimates data-centre electricity demand at about 57 TWh by FY2030.

40–45 TWh Deloitte's earlier estimate of incremental power needed by 2030 for AI data-centre demand

Cooling is not a side issue

Conceptual visual. High-density AI racks increasingly require sophisticated thermal-management systems.

The economic consequence is easy to miss. A company does not simply buy GPUs. It buys or leases a complete environment capable of feeding them with reliable electricity, removing heat, maintaining uptime, providing network bandwidth and meeting physical security requirements.

 

4  |  THE CHIP QUESTION

If compute is the engine of AI, semiconductors are the industrial supply chain behind the engine.

India's semiconductor programme is therefore adjacent to the AI bill, but it should not be counted as if every rupee were AI spending. Semicon 2.0, approved in July 2026, carries a ₹1,27,500 crore outlay and spans design, machines and materials, fabs, advanced packaging, research and talent.

₹1,27,500 crore Semicon 2.0 outlay—not an AI-only budget

The most important change is physical. India is moving beyond chip design and into assembly, testing and fabrication-related infrastructure. Micron's Sanand facility began commercial production in 2026; its first phase is planned to exceed 500,000 square feet of cleanroom space. The facility converts DRAM and NAND wafers into finished memory and storage products and is explicitly tied to demand growth from AI.

SEMICONDUCTOR PHOTO SLOT Recommended real editorial photograph: Kaynes Semicon / Micron Sanand cleanroom For publication, insert a licensed/permissioned photograph from the facility or an authorised image source.

ThePrint reported from inside Kaynes Semicon's Sanand OSAT facility, including cleanroom operators and packaging/testing equipment. Micron's official image gallery also lists Gujarat, India images and states that complete images may be reproduced subject to its stated permissions and credit requirements.

 

5  |  WHO PAYS?

This is where the headline becomes complicated. The infrastructure bill is fragmented because the infrastructure itself is fragmented.

Layer	Money / scale	What it buys	What it does NOT mean
IndiaAI Mission	₹10,371.92 cr	AI ecosystem, compute, models, skilling and related capabilities	AI-specific public mission
Semicon 2.0	₹1,27,500 cr	Design, fabs, packaging, machines/materials, R&D, talent	Broader semiconductor policy; not AI-only
Data centres	Multi-billion-dollar private market	Buildings, electrical systems, cooling, networking and land	Serves AI plus cloud/enterprise demand
Hyperscalers / cloud	Large private commitments	Cloud, AI services, data-centre capacity and related infrastructure	Commitments vary in scope; don't count as pure AI spend
Power system	Large public + private capex	Generation, transmission, substations and renewable procurement	Supports many sectors, not AI alone

Original Biz Masala graphic. Public policy outlays are shown separately because they fund different industrial layers.

The accounting trap

A ₹1.27 lakh crore semiconductor programme, a data-centre investment commitment and the IndiaAI Mission cannot simply be summed into a single 'AI investment' number. Doing so would double-count infrastructure that serves multiple sectors and turn policy budgets into a misleading headline.

 

6  |  THE FORENSIC INFRASTRUCTURE STACK

#	Layer	Why it matters
1	CHIPS	Memory, accelerators, networking silicon and processors
2	SERVERS	GPU systems, CPUs, storage and interconnects
3	DATA CENTRES	Buildings, racks, security, networking and uptime
4	POWER	Grid connection, substations, generation and backup
5	COOLING	Air, liquid cooling, water systems and heat rejection
6	NETWORKS	Fibre, interconnects, cloud links and low-latency routes
7	TALENT	Engineers, operators, researchers and chip specialists
8	CAPITAL	Public money, corporate capex, debt and institutional investment

The strategic asymmetry

India is strong in software talent and increasingly strong in digital infrastructure. It is building semiconductor packaging and manufacturing capability. But the most advanced AI accelerators remain globally sourced, and frontier-scale compute is capital-intensive. That means India's AI strategy is simultaneously a software story and a supply-chain story.

The question is not whether India can build an AI model. The harder question is whether it can build—and finance—the machine that runs millions of them.

 

BIZ MASALA FORENSIC VERDICT

India's AI ambition is real. So is the infrastructure build-out. But the 'AI infrastructure bill' is not a single number waiting to be discovered.

WHAT IS CLEAR
IndiaAI has materially expanded shared compute; data-centre capacity is scaling rapidly; semiconductor manufacturing and packaging are moving into commercial production.

WHAT IS EXPENSIVE
High-density compute requires buildings, electricity, cooling, networks and redundancy—not just GPUs.

WHAT IS EASY TO MISREAD
Semiconductor budgets, data-centre investment, power capex and cloud commitments serve multiple markets. They cannot all be labelled AI spending.

WHAT INDIA MUST SOLVE
Reliable power, affordable high-end compute, cooling, capital efficiency, supply-chain resilience and domestic capability in critical hardware.

THE BIG QUESTION
Can India become an AI superpower without first becoming an infrastructure superpower?

THE BILL IS BIGGER THAN THE AI BUDGET Because the machine behind AI spans an entire industrial economy

Murthy ASR | Biz Masala

 

REPORTING & VISUAL NOTES

This edition separates reported facts from Biz Masala analysis. Government programme outlays are not treated as equivalent to private investment. Data-centre capacity figures are identified as broader infrastructure figures rather than AI-only capacity. Forecasts and estimates are labelled as such.

Key sources consulted

·       Ministry of Electronics & IT / Press Information Bureau — IndiaAI Mission, compute capacity and AI infrastructure updates (2026).

·       Ministry of Electronics & IT / Press Information Bureau — Semicon 2.0 approval and semiconductor ecosystem updates (2026).

·       Deloitte India — TMT Predictions 2026 and AI data-centre infrastructure assessments.

·       CtrlS — Hyderabad data-centre portfolio and AI-ready infrastructure information.

·       Micron Technology — Sanand semiconductor assembly and test facility announcement and image-gallery permissions.

·       ThePrint — on-site reporting from Kaynes Semicon's Sanand OSAT facility.

·       CBRE market reporting as cited by Express Computer — 2026 Indian data-centre capacity outlook.

Image provenance / publication discipline

The CtrlS Hyderabad photograph and conceptual server-room images in this working edition were supplied in the conversation. The conceptual images are explicitly labelled as conceptual. For third-party facility photographs, Biz Masala should use images with confirmed publication rights and retain the required credit.