26/08/2026
The AI trade is becoming a power-grid trade.
The International Energy Agency says five technology giants spent more than $400 billion in 2025 and expects that total to rise another 75% in 2026. Efficiency per task keeps improving, but video, reasoning and agentic workloads can require far more energy, so aggregate demand still rises.
In the IEA base case, data-centre electricity use grows from 485 TWh in 2025 to 950 TWh in 2030, while AI-focused centres more than triple. By 2027, one refrigerator-sized advanced rack could draw roughly the same peak power as 65 households.
The constraint is physical: grids, connections, generation and capital move more slowly than compute. On-site gas can bypass connection queues, but grid-level reliability may require 30–70% extra capacity. The IEA finds the strongest AI-linked stock relationship in nuclear and grid equipment—not a blanket lift for every energy company.
Which physical constraint comes next?
MakaoApp tracks the physical constraints hidden beneath the AI trade. Stay for the next breakdown.
Sources: IEA executive summary — https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
IEA report —https://iea.blob.core.windows.net/assets/3179f7f8-01f6-4dd6-bffa-c9f7b73f1dc9/KeyQuestionsonEnergyandAI.pdf
Credits: IEA report excerpts are used under CC BY 4.0 with attribution to the International Energy Agency. Generated cover and closing images are editorial metaphors, not evidence.
Informational only. Not financial advice.