AI focus categories tracked
7
In this index
AI Workload Types
Separate training, inference, mixed-use, and general cloud facilities.
AI focus pages turn a broad infrastructure dataset into something more analytically useful by showing which facilities are oriented toward training clusters, inference fleets, or mixed workloads.
Facilities covered
295
86% of full index
Known capacity
130.1 GW
Total IT load in MW
Largest AI focus
Mixed
149 facilities
How To Read This Taxonomy
This taxonomy is one lens on a live dataset covering 344 AI data centers across 64 countries and 202 tracked operators.
What does AI focus add beyond operator and geography?
It distinguishes facilities designed for frontier-model training from those aimed at inference, cloud spillover, or mixed workloads, which changes how readers interpret hardware and power demand.
When is this slice especially useful?
Use it when tracking where dense GPU clusters may emerge, where low-latency inference capacity is being built, or how providers balance AI-specific demand with general cloud infrastructure.
Related Lenses
Cross-check this slice with adjacent taxonomies to see how geography, operator concentration, status, and power sourcing interact.
All AI focus categories
| AI focus | Facilities ↓ | Countries | Capacity | Operational | |
|---|---|---|---|---|---|
| Mixed | 149 | 44 | 61.4 GW | 101 | |
| Training | 50 | 14 | 41.7 GW | 23 | |
| Inference | 45 | 31 | 6.7 GW | 35 | |
| General Cloud | 35 | 24 | 16.2 GW | 13 | |
| HPC / Supercomputing | 10 | 9 | 2.5 GW | 5 | |
| Research | 3 | 3 | 68 MW | 3 | |
| Sovereign AI | 3 | 3 | 1.6 GW | — |
Mixed61.4 GW
149 facilities44 countries101 operational
Training41.7 GW
50 facilities14 countries23 operational
Inference6.7 GW
45 facilities31 countries35 operational
General Cloud16.2 GW
35 facilities24 countries13 operational
HPC / Supercomputing2.5 GW
10 facilities9 countries5 operational
Research68 MW
3 facilities3 countries3 operational
Sovereign AI1.6 GW
3 facilities3 countries
Frequently asked questions
- What does AI focus add beyond operator and geography?
- It distinguishes facilities designed for frontier-model training from those aimed at inference, cloud spillover, or mixed workloads, which changes how readers interpret hardware and power demand.
- When is this slice especially useful?
- Use it when tracking where dense GPU clusters may emerge, where low-latency inference capacity is being built, or how providers balance AI-specific demand with general cloud infrastructure.