# AI and digital platforms and Data centres AI models are trained and served in data centres, so AI spending turns directly into electricity demand: data centre use grew 17% in 2025 and the IEA expects it to double by 2030, with grid connections and power equipment now the binding constraint. Source page: https://www.materialinsights.org/links/ai-and-digital-platforms-and-data-centres/ [AI and digital platforms](https://www.materialinsights.org/industries/ai_platforms/) (industry) and [Data centres](https://www.materialinsights.org/energy/e_data_centres/) (energy): 4 relationships. ## How they are linked ### 1. Compute runs in data centres *Physical flow · Strong · Data centres → AI and digital platforms* Training and running AI models needs racks of accelerators with power and cooling, which only data centres provide; on the site, 6 enterprise types in AI and digital platforms depend on them, rated up to 4 of 5 for hyperscale clouds and foundation model developers. - **6** AI and digital platform enterprise types that depend on data centres - **4 of 5** criticality for hyperscale clouds and foundation model developers The dependence is highest for model developers and GPU cloud operators, whose product is compute itself; application firms that rent inference capacity are less exposed, as they can switch provider or region. Enterprise types that depend on it: - Hyperscale clouds and GPU cloud operators: Server and cooling load (criticality 4 of 5) - Foundation model developers: Training and inference compute (criticality 4 of 5) - Data licensors and labelling firms: Data storage and processing (criticality 3 of 5) - Universities and public research labs: Research compute (criticality 3 of 5) - Data, MLOps and model hosting platforms: Hosted platforms (criticality 2 of 5) - AI applications and consumer digital platforms: Inference (criticality 2 of 5) Sources: [Material Insights](https://www.materialinsights.org/industries/ai_platforms/materials/) ### 2. AI drives power demand *Transition · Rising · AI growth raises data centre electricity use* Accelerated servers draw far more power per rack than conventional servers, so each wave of AI build-out adds electricity load faster than the rest of the economy. - **about 415 TWh** data centre electricity use in 2024, about 1.5% of world consumption - **17%** growth in data centre electricity demand in 2025, against 3% for all electricity - **945 TWh** IEA projection of data centre demand in 2030 The IEA expects data centre use to double by 2030 and power use at AI-focused centres to triple. Data centres are set to drive almost half of US electricity demand growth to 2030, so the load is concentrated in a few grids rather than spread evenly. In India, data centre power demand is estimated to reach 13.56 GW by 2031 to 32, against about 1.5 GW of data centre capacity in 2025. Sources: [IEA](https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions); [S&P Global Commodity Insights](https://www.spglobal.com/commodity-insights/en/news-research/latest-news/electric-power/041025-global-data-center-power-demand-to-double-by-2030-on-ai-surge-iea); [W.Media (MeitY statement to Rajya Sabha)](https://w.media/indias-dc-capacity-almost-touched-1-5-gw-in-2025-it-ministry/) ### 3. Platform capex funds build-out *Transition · Strong · AI platform spending → data centre construction* The largest AI platforms build or lease most new data centre capacity themselves, so their capital spending plans set the pace of data centre construction. - **over US$ 400 billion** capital spending of five large technology companies, 2025 - **a further 75%** expected rise in that spending in 2026 This makes data centre demand highly concentrated in a handful of buyers. A pause in their capex guidance would slow new capacity quickly, while continued growth keeps pressure on power, chips and equipment. Sources: [IEA](https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions) ### 4. Grid and equipment limits *Shared risk · Strong · Power supply limits AI capacity* A data centre cannot run until it has a grid connection or its own generation, and the equipment for both is in short supply, so power, not chips alone, now limits AI capacity. The IEA reports that slow grid connections are delaying projects and that supply chains for gas turbines, transformers, advanced chips and IT components have tightened. Developers, mainly in the United States, are turning to on-site gas-fired generation, though many such projects are at an early stage. Sources: [IEA](https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions) ## For investors ### Key learnings 1. AI growth shows up first as electricity load: data centre demand rose 17% in 2025 against 3% for all power. 2. A few platforms' capex, over US$ 400 billion in 2025, sets data centre build rates, so their guidance is the leading indicator for the whole chain. 3. Grid connections, transformers and gas turbines are now binding constraints, so secured power, not chips alone, decides where new AI capacity can be built. ### What to track - **Hyperscaler capital expenditure guidance**: Sets the pace of data centre construction and power demand (Quarterly · company results) - **Data centre electricity demand estimates**: Shows whether AI load is tracking the doubling by 2030 (Annual · IEA) - **Grid connection queues and transformer lead times**: Signals whether power limits new capacity (Quarterly · grid operators, utility filings) ## For policymakers ### Key learnings 1. India's data centre capacity rose from 375 MW in 2020 to about 1.5 GW in 2025, and electricity demand from data centres is estimated at 13.56 GW by 2031 to 32, so power planning must allow for it. 2. Data centre load is concentrated in a few places, Mumbai holding nearly half of India's operational capacity, so local grid and water planning matter more than national totals. 3. Shared public compute, such as the 38,231 GPUs onboarded under IndiaAI, ties AI policy directly to data centre capacity and power supply. ### What to track - **Data centre capacity and power demand in India**: Shows whether generation and grid plans keep up (Annual · MeitY, Ministry of Power, CEA) - **GPUs onboarded under IndiaAI and their hourly price**: Shows how much domestic AI compute is available (Quarterly · IndiaAI Mission) ## India India's data centre capacity reached about 1.5 GW in 2025, up from 375 MW in 2020, with demand estimated at 13.56 GW by 2031 to 32; IndiaAI has onboarded 38,231 GPUs at a subsidised average of about ₹65 an hour. ## Sources 1. [AI and digital platforms: materials](https://www.materialinsights.org/industries/ai_platforms/materials/), Material Insights (2026) 2. [Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions](https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions), IEA (2026) 3. [Global data center power demand to double by 2030 on AI surge: IEA](https://www.spglobal.com/commodity-insights/en/news-research/latest-news/electric-power/041025-global-data-center-power-demand-to-double-by-2030-on-ai-surge-iea), S&P Global Commodity Insights (2025) 4. [India's DC capacity almost touched 1.5 GW in 2025: IT Ministry](https://w.media/indias-dc-capacity-almost-touched-1-5-gw-in-2025-it-ministry/), W.Media (MeitY statement to Rajya Sabha)