Affärslivet National AI Power Index 2026

Executive summary

The Affärslivet National AI Power Index ranks 49 countries on overall AI capability by combining seven dimensions from three open, authoritative sources: notable AI models and AI compute (Epoch AI), AI research output (Our World in Data / OpenAlex), and R&D intensity, researcher density, high-tech exports and internet adoption (World Bank). Each dimension is normalised 0–100 (count dimensions log-scaled to avoid a two-country wipeout), then equal-weighted. United States leads with 75.3/100; Sweden ranks 12. Unlike the paywalled Tortoise Global AI Index, every number here links to a primary dataset and every step of the method is disclosed — free to inspect, cite, download and embed under CC BY 4.0.

Key findings

The index — top economies by AI power

The Affärslivet National AI Power Index, top 15 (Sweden highlighted where shown). Each score is the equal-weighted mean of up to seven 0–100 dimensions; see the full table below for all 49 ranked economies and their per-dimension scores.

What the index reveals

The United States tops the Affärslivet National AI Power Index with 75.3/100 — the only economy strong across all seven dimensions of AI power. Its lead is built on the two hardest to replicate: frontier models and raw compute.

The single most misread statistic in the AI race: China operates MORE AI supercomputer clusters than the United States (188 vs 119), yet the US commands several times more compute — about 5× on Epoch AI's adjusted estimate (1,250,989 vs 109,411 H100-equivalents on tracked clusters, which under-count undisclosed Chinese compute). Counting data centers flatters China; measuring compute reveals an American lead. This index weights the compute, not the count.

AI power splits three ways. The United States wins the frontier (models, compute, capital). China wins volume (AI research output, cluster count). And a middle tier — South Korea, Singapore, Ireland, Denmark — competes on talent density and specialisation. No economy outside the top two is strong on compute.

Sweden ranks 12 of 49 (57.6/100), strongest on researchers/million and weakest on ai models. Like the Nordics generally, its edge is research depth, talent and digital adoption — not the compute-and-capital heavy frontier, which remains a two-country game.

Scale vs. intensity — who punches above their weight

Raw AI power rewards size: big countries build more of everything. To see who is intense rather than merely large, the chart re-ranks economies by AI compute per capita (H100-equivalents per million people). The order changes sharply — small, wealthy, compute-dense economies rise, and populous ones fall. Sweden ranks 7 of 31 on this intensity measure.

This is the difference between a country that has a lot of AI because it is big, and one that has genuinely concentrated AI capability. Both matter — but only the intensity view reveals the specialists.

By peer group — how the blocs compare

Mean AI Power score by economic bloc. Nordics leads on the average measure; the Nordics (including Sweden) sit at 58.8, ahead of or behind the EU and G7 averages depending on the bloc. Bloc averages reveal structural strength the country ranking alone hides.

Distribution & concentration — how uneven is AI power?

AI power is steeply concentrated. Across the 49 ranked economies the median score is 48.0/100 and the mean 47.6, with the middle half falling between 39.0 and 57.5. The gap between the leaders and the field is the story the average hides.

Compute is the most concentrated dimension of all: the top five economies hold about 92.4% of all tracked AI compute, and the Herfindahl–Hirschman index of compute is 6,629 — far above the 2,500 threshold economists use to flag a 'highly concentrated' market. Frontier AI is, on this measure, one of the most concentrated industries on earth.

Does R&D spending buy AI power?

Plotting each economy's R&D intensity (R&D as % of GDP) against its AI Power score reveals a very strong positive relationship (Pearson r = 0.8, n = 49). R&D investment is a foundation for AI capability — but the scatter also shows the outliers: economies that convert research spending into AI power efficiently, and those that spend heavily with less to show for it.

Correlation is not destiny. The United States and China reach the frontier through scale and capital as much as R&D ratios; several high-R&D economies rank lower because they lack the compute and capital to deploy their research.

How the index is built

Seven dimensions, three open sources, equal weight. Models and compute come from Epoch AI (notable-models and GPU-cluster databases, CC BY 4.0); AI research output from Our World in Data via OpenAlex; R&D intensity, researchers per million, high-tech exports and internet adoption from the World Bank.

Each dimension is min-max normalised to 0–100 across the countries that report it. The three count/magnitude dimensions (models, compute in H100-equivalents, publications) are log-scaled first — otherwise the US and China would crush every other country to near-zero and the index would carry no information below the top two. A country must report at least five of the seven dimensions to be ranked; the composite is the equal-weighted mean of the dimensions it reports. All dimensions are 'higher is better', so none are inverted. Equal weighting is a deliberate, transparent choice.

Sweden in focus

Sweden ranks 12 of 49 on the National AI Power Index (57.6/100), and 7 of 31 on compute intensity per capita. Its profile is the classic advanced-small-economy shape: strong on research, talent and digital adoption, weak on the compute-and-capital dimensions that only the US and China dominate.

Sweden published 2519 AI papers and hosts 4 tracked AI clusters (4,932 H100-equivalents). The Nordic bloc averages 58.8/100 — competitive with larger European economies on quality, if not on scale. For Sweden the realistic ambition is applied-AI leadership and research depth, not frontier model-building.

Full ranking — 49 economies, all seven dimensions

Every ranked economy with its composite AI Power score and its normalised 0–100 score on each dimension. One table, three authoritative open sources.

Limitations & what the data can't tell you

This index measures capability inputs and outputs that are publicly tracked — it is not a measure of AI quality, safety, or real-world economic impact. Three caveats matter. First, it rewards what is counted: closed labs disclose less, so a country's true frontier capability may exceed its visible footprint. Second, the compute dimension reflects tracked GPU clusters; undisclosed or sovereign compute is not fully captured. Third, equal weighting is a deliberate simplification — a compute-weighted index would widen the US–China lead, a talent-weighted one would favour the middle tier.

We publish the method and the full data precisely so others can reweight, extend or challenge it. That transparency is the point: this is a reproducible starting point for the debate, not the last word.

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 National AI Power Index equal-weights seven dimensions of AI capability, each min-max normalised to 0–100. Models and compute (H100-equivalents from GPU clusters): Epoch AI, CC BY 4.0. AI research (scholarly publications): Our World in Data via OpenAlex. R&D % of GDP, researchers per million, high-tech exports and internet use: World Bank. The three count dimensions are log-scaled before normalisation so the ranking carries information below the US–China duopoly. A country needs at least five of seven dimensions to rank. Scores are relative to the ranked set. This is Affärslivet's own composite — transparent and reproducible, not an official index. Nothing is estimated by us; every input traces to a named open dataset. Confidence: Medium-High.

Data dictionary

Frequently asked questions

Glossary

H100-equivalent (H100e)

A unit that normalises different AI chips to the compute of an NVIDIA H100 GPU, allowing heterogeneous clusters to be compared on one scale. ↗

AI compute

The total processing capacity available to train and run AI systems, here measured as the sum of a country's GPU-cluster capacity in H100-equivalents. ↗

Composite index

A single score built by combining several normalised indicators, used to rank entities on a concept (here, national AI power) that no single metric captures. ↗

Min-max normalisation

Rescaling values to a 0–100 range so dimensions measured in different units can be averaged; the lowest value maps to 0 and the highest to 100. ↗

Sources

• Epoch AI — Notable AI Models & GPU Clusters (CC BY 4.0)
• Our World in Data — AI scholarly publications (CC BY)
• World Bank — R&D, researchers, high-tech exports, internet use
• Stanford HAI — AI Index Report 2026
• Tortoise Global AI Index (the paid benchmark we open-source)

Related reports & data

• Countries Leading in AI: Models, Research & Investment
• AI & Tech Readiness Index
• Innovation Index
• R&D Spending by Country