What
A database and a website that link AI data centers, GPU clusters and cloud regions to the companies behind them. Every record comes with its source.
All projects
Applications from 25 Oct
Applications open · deadline 8 Nov
Applications closed
Compute Atlas
We're building an open map of the computing power behind AI: where it is, how much there is, and who owns it.
In short01 02 03
Why
The biggest AI data centers are built by a handful of companies, often for someone else: Amazon owns the New Carlisle site, Anthropic uses it. The data is public, but it's scattered, and nobody puts it together.
What you get
A real project with databases, REST APIs and web development, and a live site you can show at an interview.
Few companies, most of the power
63%
Amazon (28%), Microsoft (20%) and Google (15%) together: their share of the cloud infrastructure services market in Q2 2026.
Synergy Research, July 202695
Large AI data centers tracked by Epoch AI, with owner, main user, country, compute and power.
Epoch AI, October 2026The data is out there. We connect it.
Epoch AI already publishes great tables and maps. We don't want to copy them. We want to add three things.
- 01
Follow the ownership
We link every site and cloud region to its parent group (using GLEIF and Wikidata), so you can see how much capacity a whole group controls, not just one company.
- 02
GPUs and cloud on one map
Where the big AI data centers are, where cloud providers operate, and where the two overlap, by country and by company.
- 03
Europe and Italy
Views focused on where we live. Global datasets are still thin here: Epoch says it's expanding its European coverage.
Questions it should answer
- How much AI compute does each group control, subsidiaries included?
- Which countries host the most capacity, and who owns it?
- Who owns a data center, and who actually uses it?
- Who runs AI infrastructure in Europe and in Italy?
From scattered data to a map
- SourcesPublic datasets and APIs
- ETLScripts that fetch, clean and load the data
- DatabaseRelational, with source and date on every record
- REST APIOpen access to the data
- WebsiteSearch, profiles, map
ETL (Extract, Transform, Load): scripts that pull data from the sources into our database on a regular schedule.
First release
- AI data centers and GPU clusters from Epoch AI, linked to their parent groups
- Cloud regions of the main providers
- Search, entity and country pages
- Map
- Source and date on every record
Who we are looking for
- Database and ETL: UML, SQL, Python
- Backend: REST APIs
- Frontend: JavaScript, data visualisation
Open to all PoliMi students.
Timeline
- 25 Oct – 8 NovApplications open
- By 15 NovTeam picked, kickoff
- Nov – DecDatabase design and ETL · data in by Christmas
- Jan – FebOwnership chains and cloud regions, slower during exams
- MarchREST API and website · first release by end of March
Every record has a source
With the date it was last updated.
We say what's missing
The map shows what it knows and what it doesn't.
Open licences only
We use sources under their terms and credit them.
Data, not opinions
No judgements on companies or providers.
Want to build it?
Applications open on 25 Oct. Come back then, or follow us on Instagram so you don’t miss it.
Follow on InstagramApplications are open until 8 Nov. Pick the part you want to work on and tell us why.
ApplyApplications for this round are closed. Have an idea for another project? Propose it.
Propose a project