The Definitive Guide toAI Data Centers
Ask the GuideAboutAccount
The Definitive Guide to

AI Data Centers

A free, book-length engineering reference: 166 chapters covering the full lifecycle — siting, power, cooling, networking, financing, operations — that you can read end-to-end or enter by problem. Vendor-neutral, at practitioner depth, with every volatile number date-stamped and sourced.

166 chapters365 tables1,133 sourced figure references17 parts + appendices
Grid to GPU videosStart here: One machine

A coding agent works through a long session, and about a million tokens pass through one rack of 72 GPUs. How much energy did those tokens take from the grid, where did it go on its way through the building, and what did they actually cost?

All 37 videos

New here? The library lists the standards, references, and books this guide sits alongside — and when to reach for them instead.

Watch: Grid to GPU

37 videos · 9.4 hours

The guide, part by part

0Foundations & How to Use This GuideHow the guide works, its conventions, and where to start.▶ Video1Strategy, Workload Archetypes & EconomicsBuild vs buy vs rent, workload shapes, unit economics.▶ Video2Project Delivery, Schedule, Procurement, Contracts & RiskContracts, long-lead equipment, schedule risk.▶ Video3Site Selection, Power Procurement & PermittingInterconnection queues, tariffs, land, water, permits.▶ Video4Electrical & Energy InfrastructureGrid to busbar: substations, UPS, on-site generation.▶ Video5Cooling & Thermal ManagementAir vs liquid, direct-to-chip, heat rejection, water.▶ Video6The Building: Civil, Structural, Fire/Life-Safety & Construction ExecutionCivil, structural, fire/life-safety, build sequencing.▶ Video7Compute, Silicon & System IntegrationGPUs and accelerators, racks, integration choices.▶ Video8Networking, Fabrics & OpticsTopologies, InfiniBand vs Ethernet, optics roadmaps.▶ Video9Storage & DataCheckpointing, filesystems, data pipelines.▶ Video10Software, Orchestration & Service DeliveryScheduling, utilization, $/GPU-hr service models.▶ Video11SecurityPhysical + cyber, supply chain, confidential computing.▶ Video12Reliability, Resilience & StandardsRedundancy math, failure taxonomy, standards.▶ Video13Commissioning & Go-LiveL1–L5 commissioning, integration testing, handover.▶ Video14Day-2 Operations, Upgrades & LifecycleOperating, upgrading, depreciating the fleet.▶ Video15Sustainability & EfficiencyPUE/WUE, carbon accounting, heat reuse.▶ Video16Trends, Roadmaps & the FutureWhere the subsystems are headed.▶ VideoAAppendices & Reference DataLookup tables, standards index, reference data.

Tools

Size a cluster →
Model to megawatts: GPUs, racks, power, $/month
Check lead times →
What sets AI data center schedules right now
Compare approaches →
Air vs liquid, build vs buy, InfiniBand vs Ethernet
Test yourself →
170 questions across all 17 parts