# First collection round: use, work, costs and compute

Sources reviewed on 1 October 2026. This is a first collection round, not an exhaustive country survey. It adds 12 chart measures, 294 non-null numeric observations and three compute asset or service records. All 54 countries have four explicit review states.

## Scope and interpretation

- Adoption: Google/Ipsos online survey results for Nigeria and South Africa; Microsoft modelled generative-AI diffusion estimates for 46 countries in two periods. Keep their populations and methods separate. PwC employee survey results are pooled across five countries and must not be assigned to individual countries.
- Employment: Pnet AI-related vacancy growth is an advertisement measure, not hiring, wages, job losses or net employment. Kenyan entrepreneur research is qualitative context here; no statistically significant average result does not imply exactly zero impact. The UCT research announcement is not a completed outcome study.
- Costs: ITU 2025 data-only mobile and fixed-broadband baskets, in USD and percent of monthly GNI. Mobile coverage is 51 countries; fixed USD covers 47 and fixed GNI covers 45. The revised 5 GB mobile basket must not be spliced into the earlier 2 GB series. Source workbook hashes and cells are retained. These historical observations are not current retail quotes.
- Compute: provider-listed AWS G4dn service in Cape Town, Cassava's reported South Africa deployment, and Teraco JB4's reported completed 50 MW campus. This is a starter review, not an inventory. JB4's 30 MW addition is included in its 50 MW total. JB4 is not JB7. Cassava expansion plans are not operating sites. Unknown GPU totals, rental prices and capacity remain null.

## Consistency and provenance

`priority-evidence.json` is the canonical source for these additions. The build derives observations, compute and coverage CSVs, chart measures, and country/page evidence sections from it. Validation compares exports and chart values against the canonical records. Source definitions, periods, review dates and evidence types travel with the observations. A revision ledger records changes to observations, assets and coverage states.

The first-round observations are supplementary and not eligible to replace homepage headlines. Existing published charts, project totals and tracker-owned records remain under their existing definitions. The separate compute review must not be summed into the project pipeline.

## Gaps still to research

Firm-level adoption, use frequency, sector breakdowns, comparable wages and hires, verified device retail prices, payment checkout evidence, GPU inventory and hourly rental rates remain incomplete. Most countries still require a country-specific employment and compute search.

“No information in reviewed sources” describes only the sources checked. “Not yet researched” is a pending search. Neither means zero. Nulls are not imputed, interpolated or carried forward.