How this is built
The AI Infrastructure Map tracks $690B+ in capex across datacenters, nuclear PPAs, fusion labs, and grid plays. Here's where every data point comes from, how it's verified, and how often it's refreshed.
What's tracked
Every project on the map falls into one of five categories. Selection criteria for inclusion:
- AI datacenters / campuses — announced, under construction, or operating with confirmed capacity ≥50MW
- Nuclear PPAs & restarts — signed power purchase agreements with hyperscalers, plus restart authorisations and SMR project commits
- Private fusion companies — labs with disclosed funding, scored on team / technical / funding / timeline credibility
- Picks & shovels plays — public companies with material AI infrastructure exposure (cooling, fuel, transmission, copper, gas turbines)
- Grid stress zones — regions where the interconnection queue, reserve margin, or moratorium status is the binding constraint on AI buildout
Where the data comes from
Three layers, in order of authority:
- Primary sources — company press releases, SEC filings (8-K, 10-Q, S-1), regulatory dockets (NRC, FERC, state PUCs), official PPA announcements, court records for permitting disputes
- Industry trade press — Data Center Knowledge, The Register Datacenter, Nuclear Newswire, Canary Media, Heatmap News, TechCrunch AI
- Direct outreach — when a deal is rumoured but unconfirmed, I email the company. If there's no response, the entry is tagged "rumoured" or excluded.
How updates happen
Two pipelines run in parallel:
- Automated daily crawl — every morning at 06:00 UTC, a Claude-powered cron job pulls fresh headlines from the 6 sources above, extracts structured deal events, and writes them to the wire feed at the right of the homepage. The "Updated Xh ago" badge in the topbar reads directly from this pipeline.
- Manual verification — every major addition (large deal, new fusion company, regulatory change) is hand-checked against primary sources before it moves from the wire feed to a permanent map marker.
Scoring & subjectivity
Some things are unambiguous (capacity in MW, capex in dollars, permitting status). Others involve judgement. Where I score (e.g. fusion company conviction, picks & shovels rankings) the rubric is consistent:
- Team quality — founders' track record, key hires, advisory board
- Technical progress — verifiable milestones (gain confirmed, plasma duration, NRC approval stage), not press releases
- Funding depth — total raised, recency, lead investor quality
- Timeline credibility — degree to which announced commercial dates match independent expert estimates
Scores are recalibrated quarterly. The map shows the current snapshot.
What's NOT here
- Investment advice. The picks & shovels index is a research aid, not a recommendation.
- Speculative claims without sourcing. If you spot one, please flag it.
- Anything that requires NDA-protected information. This is built entirely from public sources.
Found something wrong?
Email hello@aiinfrastructuremap.com with the correction and a source link. The data is open, the corrections are public, and anyone who flags an error worth fixing gets a credit on the next update.