Datacenter Radar
Where the AI data-center capex goes, layer by layer.
- Status
- Live
- Launched
- 2026
- Stack
- FastAPI · React · PostgreSQL 16 · Flyway · APScheduler · DeepSeek · react-globe.gl
- Data
- Company disclosures, utility filings, trade press, 35 RSS feeds with machine classification
Live interface · captured October 2026
01What it does
Jensen Huang describes an AI factory as a five-layer cake. This app tracks six: chips, memory, optical, racks and electrical, cooling, and energy. Each layer has the companies that capture the spend, the thesis, and the constraint that decides who ships. The point is not to list suppliers — it is to show which layer is binding this quarter.
02What's inside
- 3D globe of 90 data-center sites, sized by contracted megawatts and shaded by power density
- Capex flows: hyperscaler actuals through 2025 with modelled bands to 2028
- 217 companies, including 60 private and pre-IPO names with valuation, raise history and a 69-deal supplier register
- Bottleneck register — HBM, CoWoS, transformers, turbines, interconnection queues, electricians — with severity and lead time
- News classified into moratorium, permit delay, cancellation, deals, energy, cooling, chips and supply chain
03How it works
Everything is public data, ingested on a schedule and stored in Postgres. Flyway owns the schema. The news pipeline probes every feed, drops dead ones, and runs a two-tier relevance gate so that general tech chatter stays off the board. Layer allocation of capex is modelled rather than disclosed, and the app says so on the page where the numbers appear.