Full Spectrum: AI Data Center Infrastructure

Description

Full Spectrum: AI Data Center Infrastructure 519 pages. 32 parts. 30 priced lines.

An institutional volume on the physical machine and the money behind it. Demand and whether it is real, the physics, the history, site selection and development, interconnection, generation, power equipment, cooling, water, the building, white space, security and resilience, commissioning and operations, compute, networking, the economics of training and inference, total cost of ownership, the capital stack, sustainability, regulation, the tenant market, geography, the value chain, supply chain and lead times, geopolitics, the forecast, the companies, consolidation, playbooks and scenarios. The model layer is deliberately out. A buyer who wants transformer lead times does not want agent orchestration patterns forty pages away.

The reference workbook. 51 tabs. The largest reference workbook in the library. A market ledger covering 19 named markets with the utility or operator behind each figure, an equipment design census of 55 named designs, plus cooling design, building and site, accelerated rack and depreciation policy censuses. A value chain placing 39 organizations across 13 steps. Six working calculators: capital cost builder, redundancy cost, battery replacement reserve, total cost of ownership, energy cost build up, and the revenue and occupancy bridge. Ledgers for training compute, inference price and throughput, export controls, and the capital stack in both securitizations and joint ventures. A 28 term glossary and 34 acronyms, so nothing in the volume needs looking up elsewhere.

The forecast model. 18 tabs. 2024 to 2030. A whole stack model built from 30 priced lines across three stacks: electrical equipment, building and mechanical and thermal, and information technology. Scenario driven, with a capital envelope check, a density bridge and an independent cross check that has to reconcile before the summary prints. A confidence census and a gaps register sit inside the model rather than beside it. Every driver is an input you can change, and the model recomputes.

You receive: the volume as a PDF and as a PowerPoint, the reference workbook, and the forecast model.