Affärslivet AI Compute Index

Executive summary

The Affärslivet AI Compute Index ranks 33 economies on installed AI computing power — the scarce resource that trains and runs frontier AI. It combines three dimensions from Epoch AI's GPU-cluster database: total capacity in H100-equivalents, the number of AI clusters, and their power draw, each normalised 0–100 (log-scaled) and equal-weighted. United States leads. AI compute is the most concentrated input in the field — the US alone holds about 81% of the world total — and this index makes that visible, with every figure sourced to Epoch AI and free to download under CC BY 4.0.

Key findings

The index — economies by AI compute

The Affärslivet AI Compute Index ranks 33 economies on installed AI compute — combining total capacity (in H100-equivalents), the number of AI supercomputer clusters and their power draw into one 0–100 score. United States leads; the full table below has every country and dimension.

What the data reveals

AI compute is the most concentrated resource in technology. The United States alone holds about 81% of all tracked AI compute (1,250,989 H100-equivalents); the top five economies hold roughly 92.2%. The Herfindahl–Hirschman index of compute is 6,608 — multiples above the 2,500 threshold economists use to flag a 'highly concentrated' market.

The cluster-count paradox: China operates more AI clusters than the US (188 vs 119) yet commands only about 109,411 H100-equivalents to America's 1,250,989 on tracked clusters. Counting data centers flatters China; measuring compute reveals the American lead. This index weights the compute, not the count.

A caveat on the size of the gap: these figures sum publicly tracked GPU clusters, which over-represents disclosed compute — much of it American. Epoch AI's own adjusted estimate, which accounts for undisclosed and sovereign compute (a larger share of it in China), puts the US nearer 75% of global AI compute and the US-China gap at roughly 5×, rather than the higher ratio a raw tracked-cluster sum implies. Either way the direction is identical — a clear American lead — but the exact multiple depends on how much hidden compute you assume.

Below the US, a second tier — a handful of well-funded economies and hyperscaler hosts — holds meaningful capacity, but no country outside the top two is close on absolute compute. The frontier is a duopoly with a long tail.

Compute per capita — who is dense, not just big

Absolute compute rewards size. Re-ranked by AI compute per capita (H100-equivalents per million people), the order changes sharply — small, wealthy, compute-dense economies rise. Sweden ranks 7 of 29 on this intensity measure.

Full ranking — 33 economies

Every ranked economy with its composite score and normalised 0–100 score on each dimension.

Limitations & method

This is a transparent composite: each dimension is min-max normalised to 0–100 across the countries that report it, count/magnitude dimensions are log-scaled first, and the score is the equal-weighted mean of the dimensions a country reports (minimum 2 of 3). Scores are relative to the ranked set, not absolute.

It measures what is publicly tracked — undisclosed or sovereign activity is not captured, and equal weighting is a deliberate, disclosable choice. We publish the method and full data so others can reweight or extend it. Nothing is estimated by Affärslivet; every input traces to a named open dataset.

Scoreboard (machine-readable data)

Every headline indicator with its value, period, source and confidence. Free to reuse under CC BY 4.0.

↓ CSV · ↓ JSON

Methodology & verification

The Affärslivet AI Compute Index ranks 33 economies on installed AI computing power — the scarce resource that trains and runs frontier AI. It combines three dimensions from Epoch AI's GPU-cluster database: total capacity in H100-equivalents, the number of AI clusters, and their power draw, each normalised 0–100 (log-scaled) and equal-weighted. United States leads. AI compute is the most concentrated input in the field — the US alone holds about 81% of the world total — and this index makes that visible, with every figure sourced to Epoch AI and free to download under CC BY 4.0.

Data dictionary

Frequently asked questions

Glossary

H100-equivalent (H100e)

A unit normalising different AI chips to the compute of an NVIDIA H100 GPU, so heterogeneous clusters compare on one scale. ↗

AI cluster

A group of AI accelerators networked to train or run large models together; Epoch AI tracks hundreds worldwide. ↗

Herfindahl–Hirschman index (HHI)

A standard measure of market concentration; above 2,500 is considered 'highly concentrated'. ↗

Sources

• Epoch AI — GPU Clusters & Data Centers (CC BY 4.0)
• World Bank — population (per-capita)

Related reports & data

• National AI Power Index
• The Biggest AI Data Centers
• Countries Leading in AI