Guide › Part 1
Part 1
Strategy, Workload Archetypes & Economics
8 chapters
1.11.21.31.41.51.61.71.8
The Archetype Decision Framework: Workload Is the Master Variable
The workload sets density, cooling, fabric, redundancy and siting; your rack, measured traffic and uptime target size each. Fix the archetype first: a wrong site means moving, short utility capacity a new service request, a weak floor structural rework, wrong power or cooling plant a tear-out under a running hall.
Training Data Centers: Synchronous, Dense, Checkpointable
A training cluster is one synchronous supercomputer that moves at its slowest GPU's pace. Design for goodput per megawatt, dense and checkpointable, with cooling and fabric chosen from your rack's specs and measured traffic, before you cut steel.
Inference Data Centers: Bursty, Distributed, Always-On
An inference data center serves many independent requests against a latency SLO. Size it to goodput per dollar and tokens per watt at real batch sizes, not to peak training throughput; designing it as a de-rated training cluster gets nearly every subsystem choice wrong.
Post-Training, Fine-Tuning & RL: The Hybrid Middle
Classify post-training before sizing: supervised and offline preference updates have a fine-tuning profile, while rollout-heavy RL combines generation and policy updates. Size each phase from measured work and accepted learning throughput, then choose shared or separate GPU pools.
Edge Inference & Distributed Micro-Datacenters
Edge inference trades cheap power and consolidation for proximity and local control. It pays when a centralized region fails a latency, data-boundary, WAN-autonomy or input-transport-cost requirement.
Procurement Archetypes: Build vs Buy vs Rent
Match capacity to the durability of your demand: compare ownership for durable, contracted load and rental for uncertain load, valuing only the exit option the contract actually grants.
The Requirements-and-Consequences Matrix
Name the workload profile, then turn its measured demands into a signed design basis. Cooling, fabric, storage, redundancy and siting each need their own evidence about the equipment, facility, traffic and service level. The workload label alone selects none of them.
Business Models, Economics & ROI
Four numbers decide an AI data center's return: capex per watt, assumed economic earning life, billable utilization and post-deflation price. Get any one wrong and the asset strands. Model book depreciation and debt service separately.