Data Centres 2005-2025: $500bn Investment — The New Energy Customer

Sep 25

3 min read

1 September 2026

Data Centres: $500bn Investment — The New Energy Customer

Key Points

Investment nearly doubled in two years → $500bn in 2024
For Comparison: Upstream Oil & Gas — $570bn
Electricity Demand (2024–2030): 415 → 945 TWh
Installed Capacity (2005–2025): 21 → 114 GW
Growth Drivers: IT → Cloud → AI
Power Package Scale-Up: 26 → 200 → 2,000 → 5,000 MW

How Data Centres Are Powered

Coal — 30%
Renewables — 27%
Natural Gas — 26%
Nuclear — 15%

Infrastructure Constraints

Chips
Grid
Fibre
Water and Land

AI Chips

Design: NVIDIA (🇺🇸), AMD (🇺🇸)
Manufacturing: TSMC (🇹🇼), Samsung (🇰🇷), Intel (🇺🇸)
High-Bandwidth Memory: SK hynix (🇰🇷), Samsung (🇰🇷), Micron (🇺🇸)

What Data Centres Deliver

AI Model Training — close to power
AI Inference — close to users
Speed-Sensitive Processing — trading and gaming
Cloud Computing
Data Storage

Scale Example — ChatGPT: more than 2.5 billion messages per day worldwide, as reported by OpenAI in July 2025.

Technology Milestones

🇺🇸 2006 — Public Cloud: Amazon Web Services launches EC2.

🇺🇸 2007–2011 — Cloud Growth: AWS, Google Cloud and Microsoft Azure expand global cloud infrastructure.

🇨🇦 🇺🇸 2012 — Deep Learning: University of Toronto develops AlexNet using NVIDIA GPUs. Geoffrey Hinton, Alex Krizhevsky and Ilya Sutskever demonstrate the potential of GPU-powered neural networks.

🇺🇸 2013–2016 — Hyperscale: Amazon, Microsoft, Google and Meta expand large, modular data-centre campuses globally.

🇺🇸 2017 — Transformer: Google Research publishes Attention Is All You Need, establishing the foundation for modern large language models.

🇺🇸 2018–2020 — GPT-3: OpenAI, Microsoft and NVIDIA drive rapid growth in model size, training compute and GPU infrastructure.

🇺🇸 2021–2022 — ChatGPT: OpenAI, Microsoft Azure and NVIDIA bring generative AI into mass-market use.

🇺🇸 2023–2024 — Generative AI: Microsoft/OpenAI, Google, Meta, Amazon, Anthropic and NVIDIA accelerate investment in AI models, GPU clusters and data centres.

🇺🇸 🇨🇳 2025 — AI Race: OpenAI/Stargate, Microsoft, Meta, Google, Amazon, xAI and DeepSeek compete in model capability, chips, computing capacity, power and grid access.

Methodology

Chart Basis: The 2005 global installed-capacity baseline is 21.4 GW. When the additions shown in the chart are combined with this baseline, they produce the estimated 2025 total of approximately 114 GW.

Global Power Capacity: Capacity includes IT equipment, cooling and auxiliary systems.
Capacity Additions: Each bar represents the net addition to global installed data-centre capacity during the labelled year or period, not total installed capacity or a cumulative running total. Where several years are grouped, the bar shows the cumulative addition over the complete period. Because the periods differ in duration, the bars should not be interpreted as directly comparable annual growth rates.
Selected Technology Milestones: Dates, labels and country flags identify selected technological and commercial developments associated with the expansion of public cloud, hyperscale infrastructure, deep learning, transformer models and generative AI.
Attribution: The milestones provide historical context and identify developments associated with increasing computing demand and infrastructure scale. They should not be interpreted as the sole causes of capacity additions during the corresponding periods.
Country Flags: Flags indicate the principal geographic origin of the selected technology, company or institution associated with each milestone. Their size, position and proportion within a bar do not represent national or regional shares of global capacity additions.
Electricity Demand: Annual electricity consumed by data centres, measured in TWh.
Power Packages: One published MW figure is used per facility or campus, without separating IT load, cooling, PUE or auxiliary consumption.
Project Status: Projects may be operating, under construction or stated as future capacity targets. The applicable status is identified in the underlying project description.
Energy Mix: The energy mix represents the approximate global electricity mix supplying data centres in 2024.
Investment Comparison: Data centres — approximately $500bn; upstream oil and gas — approximately $570bn. Both figures relate to 2024 but come from separate IEA datasets. The comparison illustrates their relative investment scale; the underlying investment categories are not identical.
Investment Horizon: 2024 is retained as the latest directly comparable year. Later IT CAPEX and hyperscaler CAPEX forecasts are excluded because they use different scopes.

Sources

Investment: IEA, Energy and AI — data centres, $500bn in 2024; IEA, World Energy Investment 2024 — upstream oil and gas, $570bn in 2024.
Capacity, Demand and Energy Mix: IEA, Energy and AI; IEA Energy and AI Observatory; Visual Capitalist reproduction of the IEA capacity series.
Technology Milestones: AWS; University of Toronto; Google Research; OpenAI; Microsoft; NVIDIA; company and institutional publications.
AI Chips: NVIDIA; AMD; TSMC; Samsung; Intel; SK hynix; Micron; semiconductor-industry publications.
AI Usage: OpenAI reporting.
Power Packages and Projects: Start Campus; OpenAI/Stargate; Meta; project and operator announcements.