Release 2026.2 Updated 17 Jul 2026

National AI compute GBR

The National AI Compute Index measures the amount of reliable, affordable, accessible, and strategically controlled AI-processing capacity available to a country's startups, researchers, enterprises, and public institutions.

Effective AI compute
per 2023 data
Raw AI computeHardware capacity: accelerators, clusters, data centres76.7
EnablementShare of raw capacity that is actually usable82.7%
Effective AI computeRaw × enablement · Strong63.4
Enablement axes
do the chips become usable national capacity
AxisWhat it asksScore
AccessibleCloud access and who may use it91.3
AffordableCost relative to local ability to pay93.5
ReliablePower, uptime, and supply continuity100.0
Strategically controlledSovereign control of the stack50.0
TalentPeople to build and operate it100.0
SoftwareStack and developer ecosystem75.0
Inputs
AI accelerator / cluster tier (0–3)2
Data-center tier (0–3)3
Hyperscaler cloud regions7
National compute programmeyes
Supply-continuity tier (1 best–3 restricted)1
Strategic-control tier (0 foreign–3 sovereign)1
Talent tier (0–3)3
Software-stack tier (0–3)2
GNI per capita, PPP$58,140
Access to electricity (%)100.0
How this is computed

AI compute is not chips alone; it is the practical ability to turn electricity, data, hardware, software, and talent into training and inference capacity. So the index measures two levels. Raw AI compute is the hardware that physically exists, accelerators, clusters, and data centres. Effective AI compute is that raw capacity conditioned by whether it can actually be used: accessible, affordable, reliable, strategically controlled, with talent and software. Effective equals raw times enablement, so idle or unusable hardware scores low. Income and electricity are live World Bank data; capacity, supply, control, talent, and software inputs are curated and openly listed, estimated from development tier where a country is not yet curated, and overridable by an authorised administrator. Separate from the Optional Work Index.