The AI-native signal. Herfindahl–Hirschman Index of frontier-model training compute by organisation, per year (0–10000; higher = more concentrated). Global, not per-country — frontier compute lives in a handful of labs.
| Country | Year | Labour Share % | Top 1% Wealth % | Top 10% Wealth % | Top 1% Income % | CCI |
|---|
From the World Inequality Database (WID.world), labour share of GDP at factor cost (series ylsgdp). Capital share — the CCI — is its complement: CCI = 100 − labour share. A rising CCI means a larger slice of national income accrues to capital rather than wages.
Top 1% / top 10% wealth share (shweal992j) and top 1% pre-tax income share (sptinc992j), equal-split adults, from WID.world. These measure how concentrated the captured surplus becomes once it leaves the wage bill.
Herfindahl–Hirschman Index of frontier-model training compute by organisation, per year, computed from Epoch AI's frontier-models dataset (training compute in FLOP, attributed to the lead organisation). 0–10000; higher = fewer labs control the means of producing intelligence. This is a global series — frontier compute does not distribute by country.
Resource rents (Norway, Saudi Arabia): both read with very high capital share because petroleum income books as a return to capital, not labour. Norway's CCI ≈ 62 and Saudi Arabia's ≈ 84 reflect resource dependency, not AI-driven capture — the same distortion that inflates Norway's corporate-tax share in the IC Index. Read their trend, not their level.
Provisional latest year (*): WID nowcasts the most recent year or two; rows marked * are extrapolated and will be revised as final national-accounts data arrives.
Reading the index: the factor-share trend is a slow, decades-long backdrop driven by globalisation, automation, and policy long before AI. It is the curve the intelligence thesis predicts will bend; the compute-concentration panel is where that bend, if it comes, should show first.