TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Ethics & Provenance · AGE 18—22 · POSITIONAL · HELD
The data centre is a building
Siting, cooling, water and heat recovery are design decisions, and the brief now reaches architects.
Our position
The infrastructure that runs AI is a building type. It has a site, a roof, a power connection that can rival a town's, cooling that in many designs draws water, and a brief that more and more often reaches an architect. The profession taught itself to read embodied carbon in concrete and steel. It cannot look away now that the carbon has moved into the server hall. This is the same climate responsibility, in a new building type.
Why we hold it
The IEA's 2025 report treats data-centre electricity demand as a planning problem for grids and cities. That is a problem architects know how to read. Siting, cooling, water, heat recovery and the fabric around the racks are design decisions, and a profession that turns them down leaves them to be made badly.
The strongest objection
A data centre is an engineering product in a shed. The decisions that matter — chip efficiency, cooling technology, grid mix — belong to operators, engineers and regulators. The architect's share is a façade and a car park. Calling the server hall "architecture" inflates the discipline's relevance and distracts students from the buildings they will actually design.
What would make us revise it
If data-centre briefs stop reaching architects — operators moving wholly to prefabricated modules designed in-house — or if grid decarbonisation and cooling advances make the energy and water question marginal, the card is demoted from a position to a footnote. Reviewed every edition against what is actually being commissioned.
Try it
Find one data centre built or announced in India in the last three years. Describe it as you would any building type: site, power source, cooling method, water, what stands next to it. Then write the one question you would ask its client first.
Take it to crit
Does the student treat AI's physical infrastructure as architecture they could be asked to design — with siting, cooling and water as design decisions — or as someone else's distant problem?
How it works
A data centre's energy goes into computing and into removing the heat that computing makes. Cooling can be air, water, or liquid to the chip, and each choice has a consequence on site: a water draw that ranges from almost nothing in an air-cooled hall to very large, noise, exhaust heat, the possibility of heating a neighbourhood with it. The studio already has words for all of this: embodied and operational carbon, orientation, thermal mass, services. The Generation is not free card on this map is the same fact seen from the prompt end.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
- The Hidden Layers
- hidden-layers-compute-energy-labour-bias
- IEA, Energy and AI, 2025
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.