Compute Atlas

Every AI answer ends in a building. A 20 MW AI data centre needs a 26 MW grid connection, about 2 lakh litres of water a day, flat land, fibre, and an energisation date. In India, the facts that decide whether a site can deliver these are scattered across public APIs and private documents, and the private documents often disagree. A broker brief says 30 MW; the utility's letter says 12 MW, conditional. Today an analyst reconciles this in their head, and land gets bought before power is proven.

The people who feel this are site-selection and data-centre engineering advisory teams, developers' expansion teams, and investors' technical diligence teams. They screen three to six candidate sites per project and must decide which one deserves the next round of costly diligence, and what could invalidate it.

Compute Atlas is a map-first workspace for screening Indian AI data-centre sites. You draw or import a parcel anywhere in India, choose a workload (a 20 MW campus or a 5 MW modular unit, every assumption editable), and each requirement gets one of four verdicts: supported, not met, unknown, or conflict. Missing evidence is never coerced into a pass. Two sources that disagree are shown side by side with paragraph references, and no winner is picked. There is no overall score anywhere in the product.

Real recorded data drives the context: 804 substations and 2,226 power lines from OpenStreetMap, 203 carrier facilities from PeeringDB, NASA POWER climate normals, and state boundaries. You can ingest your own PDFs or text; claims like "18 MW" or "connection by March 2028" are extracted with their paragraph and weighed like any other evidence.