Inside the Data Center

What it does

Inside the Data Center is a four-layer interactive explorer for the AI build-out's physical supply chain.

Layer 0 (World). A dark world map with glowing dots at the major data center clusters: Northern Virginia, Frankfurt, Singapore, Mumbai, Dallas, Phoenix, Tokyo, and more. Dot size is proportional to cluster capacity.

Layer 1 (Anatomy). Click any cluster and the map fades back. A reference floor plan of a typical hyperscale facility appears with six labeled sections: Server Halls, Electrical Room, Cooling Plant, Generator Yard, Battery Room, and Meet-Me Room. Each section is colored and clickable.

Layer 2 (Parts). Click any section and you get the parts inside it. The Electrical Room shows HV transformers, MV switchgear, UPS systems, and automatic transfer switches. Each part card shows what it does, its HS trade code, lead time, and typical OEMs.

Layer 3 (Origins). Click any part and you go back to the world map, but now with arcs animating from the top origin countries to the cluster you started at. Arc thickness is proportional to share of US imports for that part's HS code. A side panel lists the major OEMs and their headquarters countries.

What this is not

This is not a real bill of materials for any specific facility. Real BOMs are private. This is a reference architecture compiled from publicly available sources: Open Compute Project specifications, Uptime Institute Tier whitepapers, ASHRAE TC 9.9 publications, US import data from UN Comtrade and USITC DataWeb, OEM earnings calls via SEC EDGAR, and industry trade press.

Origin shares are illustrative aggregates of US import data by HS code. HS codes are coarse, especially for categories like cooling and networking gear that include non-data-center products. Treat numbers as directional, not authoritative.

Why this exists

Most data center visualizations on social media show locations as bubbles on a map. That story is saturated.

The interesting story is what's inside the building and where every part of it comes from. Transformers backed up to 18-30 month lead times. UPS batteries dominantly Chinese lithium-ion. Optical transceivers concentrated in three Chinese vendors. Cooling chillers split between US and Asian manufacturers. Diesel generators heavily American with multi-year backlogs.

This build makes the supply chain navigable. You don't need an analyst report or a paid intelligence subscription. You need a few public datasets, a map, and a willingness to click around.

Data sources

All public, all free.

• UN Comtrade for international trade flows by HS code
• USITC DataWeb for US-specific HS-10 import data
• SEC EDGAR for OEM 10-Q backlog and lead-time commentary (Eaton, Hubbell, GE Vernova, ABB ADRs, Caterpillar, Cummins)
• Open Compute Project specifications for reference architecture
• Uptime Institute Tier whitepapers
• EPA GHG Reporting Program facility-level data
• Wikipedia "List of data centers" for cluster cross-reference
• Hyperscaler press releases for new site announcements

How to read it

Start at the world view. Pick a cluster you have a reason to care about. Northern Virginia if you want the AI epicenter, Mumbai if you want the fastest-growing one, Frankfurt if you want the EU hub.

Drill into the Electrical Room. Click the HV transformer. Watch the arcs. Notice that almost all US transformer imports come from Mexico, Korea, Germany, and Turkey, with India a small but growing share.

Now do the same for UPS battery banks. Notice the Chinese share is over half. That's the lithium-ion supply concentration story.

Now compare diesel gensets. Notice they're heavily US-domestic, but the lead time is 12-24 months. Different shape of bottleneck.

Six sections, twenty-one parts, twelve origin countries each. There's enough to wander for a while.

Tech stack

Next.js App Router, Mapbox GL JS for the maps and arc layer, hand-written isometric SVG for the anatomy view, React state machine for the four layers. All data pre-fetched and stored as JSON in public/data/inside-dc/ so there are no live API costs in production.