Artificial intelligence is generally encountered as language appearing on a luminous surface, yet its actual form extends through processor halls, substations, fibre routes, cooling towers, backup generators, water circuits, replacement components and territories capable of absorbing continuous demand. Intelligence at scale must be refrigerated, and this ordinary technical requirement is beginning to shape the geography of computation as decisively as access to capital or specialised labour. The expansion of data centres is therefore not simply a story of faster models but of siting, grid capacity, thermal management, emissions and local planning. Springer and Hasanbeigi show that the consequences vary sharply by region because facilities enter electrical systems with different generation mixes, regulatory conditions and growth limits; Abera and Chen demonstrate that compute scheduling and cooling cannot be treated as separate technical problems when thermal behaviour affects both performance and total energy consumption; He and Qu push the question further by examining whether low-grade thermal resources can displace electricity-intensive cooling. Together these positions move the debate away from the abstract footprint toward an infrastructural description of what computation requires in a particular place. A facility cooled with potable water in a drought-prone basin is not equivalent to one using reclaimed water in a cooler climate, just as an annual renewable-energy contract is not equivalent to continuous low-carbon supply at the hour and location where processing occurs. HeatSentence enters at the moment when an apparently universal capacity receives a climatic address and its hidden mechanical conditions become public facts. The most useful systems may consequently be judged not only by latency or model performance but by the intelligence of their territorial arrangements: whether workloads can move toward periods of abundant electricity, whether waste heat can support nearby housing or industry, whether cooling is adapted to climate, whether water demand is disclosed, and whether communities receive durable benefits rather than temporary construction followed by enclosed technical occupation. None of this requires describing artificial intelligence as an ecological catastrophe or abandoning computational research. It requires replacing the fiction of placeless intelligence with a precise account of the apparatus that sustains it. Computation is becoming one of the architectures through which energy systems, municipal planning and regional development are reorganised. Refrigeration is not a hidden service beneath intelligence; it is the climatic condition through which intelligence becomes continuously available.
Springer, C. and Hasanbeigi, A. (2025). Data Centers in the AI Era: Energy and Emissions Impacts in the U.S. and Key States. Global Efficiency Intelligence.
Abera, N. B. and Chen, Y. (2026). Coordinated Cooling and Compute Management for AI Datacenters. IEEE Transactions on Cloud Computing.
He, Q. and Qu, C. (2025). Waste-to-Energy-Coupled AI Data Centers: Cooling Efficiency and Grid Resilience.
Crawford, K. (2021). Atlas of AI. Yale University Press.
