Material Insights · Understanding the Material World

AI and digital platforms and Data centres and cloud

AI models are trained and served in data centres, so AI spending turns directly into data-centre construction and power demand: five large tech firms spent over USD 400 billion of capex in 2025, data-centre electricity use rose 17%, and power and equipment are now the bottleneck.

AI and digital platforms (industry) and Data centres and cloud (industry): 5 relationships.

How they are linked

1. Buys from

Physical flow · Strong · AI and digital platforms buy from data centres and cloud

AI and platform firms rent compute, storage and networking from cloud providers or colocation sites, or build their own, so a large share of their costs is data-centre capacity.

  • ₹12.8 of every ₹100 of AI and digital platforms output, all rounds (India 2022)
  • ₹10.8 bought directly

Most of this is bought directly rather than through other suppliers. The 2022 table predates most of the generative AI build-out, and data centres that the largest platforms own count as their own investment rather than as a purchased input, so the table understates how much the industry relies on data-centre capacity.

Sources: OECD

2. Compute capacity

Physical flow · Strong · Data centres to AI model developers and clouds

Training and running large models needs GPU clusters with dense power and cooling, which only purpose-built data halls provide; the site rates data centres 4 of 5 for hyperscale and GPU clouds and for foundation model developers.

  • 4 enterprise types in AI and digital platforms that depend on data centres
  • 4 of 5 criticality for hyperscale clouds and foundation model developers

Data, model hosting platforms and consumer AI apps depend less, rated 2 of 5, because they mainly need cloud hosting and inference capacity that many providers can sell. The firms rated 4 of 5 need buildings, power and cooling at scale.

Enterprise types that depend on it:

  • Hyperscale clouds and GPU cloud operators: Buildings, power and cooling (criticality 4 of 5)
  • Foundation model developers: Compute capacity (criticality 4 of 5)
  • Data, MLOps and model hosting platforms: Cloud hosting (criticality 2 of 5)
  • AI applications and consumer digital platforms: Inference capacity (criticality 2 of 5)

Sources: Material Insights

3. AI capex builds data centres

Transition · Rising · AI spending drives data-centre construction

Large platform companies are racing to add AI compute, and almost all of that spending goes into data centres, servers and the power to run them.

  • more than USD 400 billion capex of five large technology companies, 2025
  • 75% further rise in that capex expected in 2026
  • USD 125 to 145 billion Meta's 2026 capex guidance, raised from USD 115 to 135 billion by May 2026

The spending is concentrated in a handful of firms, so data-centre builders and their suppliers depend on the capex plans of a few customers. Meta cited higher component prices and extra data-centre costs when it raised its 2026 guidance; if AI revenue fails to justify the outlay, capex cuts would reach data-centre demand quickly.

Sources: IEA; Yahoo Finance

4. Power demand and bottlenecks

Transition · Rising · AI workloads raise data-centre power use

AI training and inference run on accelerator clusters that draw large amounts of power, so AI growth shows up as rising data-centre electricity demand and pressure on grids, turbines and transformers.

  • 17% rise in data-centre electricity demand, 2025
  • 3% growth of world electricity demand, 2025
  • double data-centre electricity use by 2030, with AI-focused sites tripling (IEA)

AI-focused data centres grew even faster than the 17% average. The IEA reports shortages of gas turbines and transformers, along with slow approvals, holding up grid connections, so power availability now shapes where and how fast AI capacity can be built.

Sources: IEA

5. India's AI compute and data centres

Policy · High in India · Indian policy funds shared AI compute in Indian data centres

India's government offers start-ups and researchers affordable access to a shared pool of GPUs onboarded through the IndiaAI compute portal, and data-centre operators have made large investment commitments in the country.

  • more than 38,000 GPUs onboarded on the IndiaAI compute portal, April 2026
  • ₹10,372 crore IndiaAI Mission outlay
  • about USD 173 billion data-centre investment commitments in India since 2021

India's operational data-centre capacity reached about 1.75 GW in H1 2026 and was on track for 2 GW by year end, with Mumbai taking 94% of new H1 supply. More than 15% of commitments since 2021 were tied to AI, so AI demand is a growing but not yet dominant driver.

Sources: National e-Governance Division, Government of India; CXOToday (CBRE data)

For investors

Key learnings

  1. AI capex is data-centre demand: five large tech firms spent over USD 400 billion in 2025, with a further 75% rise expected in 2026.
  2. Power is the binding constraint: data-centre electricity use rose 17% in 2025 against 3% for the world, and turbines and transformers are scarce.
  3. Demand is concentrated in a few buyers, so data-centre builders carry the risk that AI returns disappoint and capex is cut.

What to track

  • Hyperscaler capex guidance: Leading indicator of data-centre construction and equipment orders (Quarterly · company results)
  • Data-centre grid connection queues and power prices: Show whether power limits AI capacity growth (Quarterly · utilities, grid operators, IEA)
  • AI revenue of platform firms: Shows whether the capex is earning a return (Quarterly · company results)

For policymakers

Key learnings

  1. India's data-centre capacity reached about 1.75 GW in H1 2026, with about USD 173 billion of commitments since 2021, so power and land policy shape its AI capacity.
  2. The IndiaAI Mission had onboarded more than 38,000 GPUs to its common compute portal by April 2026, giving domestic users affordable compute through existing data-centre capacity.
  3. Data centres doubling their electricity use by 2030 makes grid planning and renewable supply central to AI policy.

What to track

  • Operational data-centre capacity by city: Shows concentration risk in Mumbai and the spread to other states (Half-yearly · CBRE, industry reports)
  • GPUs available and used on the IndiaAI portal: Shows uptake of public AI compute (Quarterly · Ministry of Electronics and IT)

India

India had about 1.75 GW of operational data-centre capacity in H1 2026 and about USD 173 billion of commitments since 2021; the ₹10,372 crore IndiaAI Mission had onboarded more than 38,000 GPUs for shared use by April 2026.

Sources

  1. OECD Input-Output Tables 2025 edition, India 2022, OECD
  2. AI and digital platforms: materials, Material Insights (2026)
  3. Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions, IEA (2026)
  4. Hyperscalers hit USD 700 billion in 2026 AI spending plans, Yahoo Finance (2026)
  5. IndiaAI Mission expands AI ecosystem with affordable compute and startup support, National e-Governance Division, Government of India (2026)
  6. CBRE H1 2026 report: India data center capacity hits 1.75 GW, set to cross 2 GW by year-end, CXOToday (CBRE data) (2026)