★
Menu
Accueil Comment c'est construit Watchlist La pile
Matériel PucesComparer les puces Serveurs & systèmesRéseau MémoireFonderies & Packaging Refroidissement
Infrastructures Data centersAlimentation ProjetsCloud et calcul GPU
Intelligence Modèles IALogiciels Produits IABenchmarks Articles de rechercheJeux de données Concepts
Écosystème EntreprisesActions PersonnesChronologie TendancesAvenir de l'IA
Formation Actualités Interroger les données Agents IA
À propos À propos de nous Méthodologie Données & API Avertissement
Options
DONNÉES & API

Accès JSON gratuit en lecture seule aux données en cache du site.

Mode sombre

Vue guidée
Nouveau dans l'univers des puces, des data centers et du jargon des modèles ? Nous expliquons les termes clés au fil de votre navigation, en langage clair. Les mêmes données, avec l'aide intégrée.

Vue expert
Vous connaissez déjà le sujet. Juste les données : claires, rapides et concises, sans explications superflues. Il s'agit de la vue par défaut.

Langue de l'interface
‹ Retour
Lecture rapide 6 min de lecture · ← Tous les articles explicatifs

AI campuses do not land randomly. Every siting decision is a compromise between five resources, and once you know them you can look at a map and predict where the next campus goes — the way analysts and land brokers already do.

1. Power first — everything else is negotiable

A gigawatt campus needs transmission-scale electricity, so sites cluster where the grid is strong and the interconnection queue is short: next to existing substations, along high-voltage corridors, at retired coal or industrial sites whose grid connection outlives them (Homer City is the template), or at the fence line of a power plant itself (Susquehanna). Training doesn't care about latency to users, which freed the industry from metro areas — that is why frontier compute rose in Abilene, New Carlisle, Richland Parish and North Dakota rather than Silicon Valley.

2. Water — the old tell, now changing

Evaporative cooling made big campuses river-and-lake creatures — Google's Council Bluffs sits by the Missouri; Data Center Alley grew along Virginia water and fiber. So "follow the water" long worked as a scouting rule. It is weakening: closed-loop liquid designs like Fairwater fill their loops once and sip thereafter, precisely because water optics became a community flashpoint. Where you still see big water rights being acquired, expect evaporative economics; where you see dry coolers on the renders, expect closed loops that traded water for a little more electricity.

3. Fiber — the quiet constraint

Campuses must sit on or near long-haul fiber backbones (and for multi-site training, on routes to their sibling campuses — Microsoft's AI WAN links its Fairwater sites). Old railroad and highway rights-of-way carry much of America's fiber, which is why data-center corridors often shadow them. Corning's fiber and Ciena's coherent optics are the layer that makes a remote gigawatt site usable.

4. Climate, land and the local bargain

Cool, dry climates give free cooling hours and better PUE; flat cheap land lets you build the next hall beside the last; and the political package — tax abatements, permitting speed, a utility willing to build generation (Entergy for Meta-style) — often decides between finalist sites. The counterweight is the new politics of load: who pays for grid upgrades, air permits for on-site turbines (Memphis), and rising retail rates.

The scout's checklist

When a big land purchase makes local news, check: distance to a 345 kV line or substation; a water story (river rights or dry-cooler design); fiber routes; who the utility is and what they just asked regulators for; and any turbine or transformer orders reported nearby. Two or more of those tells together usually mean a campus — the records in our Data Centers and Power sections each list exactly these ingredients, so you can study the pattern on the real examples.

Plus dans Systèmes et infrastructures

Inside an AI Data Center Lire → Networks: How 100,000 GPUs Become One Computer Lire → Liquid Cooling: Why AI Got Wet Lire → Neoclouds: The GPU Landlords Lire →