One thing that stands out reading through the IEA’s recent report, “Key Questions on Energy and AI” (link below), is the growing mismatch between the speed of software evolution and the speed of physical infrastructure.
AI capabilities are scaling extraordinarily quickly, but electricity grids, planning systems, permitting, network infrastructure and urban deployment models evolve far more slowly.
The challenge is not simply AI itself. On that score, the report makes an important point: efficiency improvements in AI are happening at remarkable speed, yet overall energy demand continues to rise because the scale and complexity of AI applications are expanding even faster.
That raises an interesting systems question. There is a strong argument that the future is not only about building larger centralised data centres, but also about designing more intelligent, energy-efficient distributed infrastructure at the edge of the network itself:
- AI-assisted network optimisation
- edge compute closer to the point of use
- more efficient urban infrastructure
- intelligent load balancing
- distributed sensing and operational systems
- flexible, resilient city-scale networks
In many ways, this becomes an infrastructure design challenge as much as a computing challenge.
The more intelligently — and robustly — we design the physical layer of cities and networks, the less unnecessary energy, latency and operational overhead we may ultimately push back into the core.
The IEA report is largely focused on data centres — understandably so — but it also indirectly points toward something broader:
AI, energy systems, connectivity and urban infrastructure are now becoming part of the same conversation.
Link to IEA Report:
https://iea.blob.core.windows.net/assets/3179f7f8-01f6-4dd6-bffa-c9f7b73f1dc9/KeyQuestionsonEnergyandAI.pdf