How the figures are compiled

An earlier version of this site published a larger set of metrics across the same 7 pillars. Most of those figures were plausible, but they had not been verified against the publisher, and we removed them.

We publish a metric only if its figure is taken from a primary source and another person can check it against that source. In practice, this rule excludes the following measures, which an index of national AI capacity would usually include.

• Datacentre capacity and energy useWe have not found a public datacentre register covering Cyprus that we could use. The figures that have been published for Cyprus are based on estimates.
• Number of AI companies and AI investmentThe NACE classification of economic activities has no code for developing AI, so the Registrar of Companies cannot identify AI companies. Commercial trackers that report national totals for AI investment do not publish their methodology, so they do not meet the source rule.
• AI research publications by countryTo attribute a paper to a country, each author’s affiliation has to be identified. Affiliation records in the open bibliographic databases are too inconsistent to support a national count unless they are first cleaned and matched, and we have not done that work. The list of recent publications on the overview page draws on the same data, but it shows titles only and gives no count.

Where few figures meet this rule, the pillar has few metrics. We do not add unverified figures to make up the number.

Every figure carries two separate labels: a quality tier and a freshness state. The two are independent, so a figure can be current and still be an estimate.

Freshness is calculated each time a page is produced, from the date of the latest observation and how often the source publishes. It is not stored, so a figure cannot remain marked as current once its update is overdue. A figure that we compile by hand from a document is dated by its review rather than its reference period, because a count taken from a register is correct only as of the day the register was read.

8 of the 14 metrics can be updated automatically from Eurostat. When automatic updating is switched on, new data from the source replace the last reviewed figure only if they are complete and at least as recent. Otherwise the reviewed figure is kept, so a source that returns an incomplete series cannot shorten a published chart without notice.

Every metric page and every API response states which of the two is shown: an automatic update from the source, or the last figure we reviewed and published. A reviewed figure is the value the source gave as of the review date shown beside it.

Each metric is converted to a score from 0 to 100 against an absolute reference band, that is, a fixed floor and ceiling. Metric scores are combined into a pillar score by weighted average, and pillar scores are combined in the same way into the overall index score. Scores summarise the published values, and the values take precedence.

Scoring method: Absolute reference bands · version 1.0.0 · in force from 2026-08-07

higher is better   100 × (value − floor) / (ceiling − floor)
lower is better    100 × (ceiling − value) / (ceiling − floor)
limited to 0–100, then weighted and averaged

We use absolute bands rather than bands based on the peer group. If each score were scaled between the lowest and highest values in the peer group, a country’s score would change whenever another country’s figure changed, and changes from one year to the next could not be interpreted. Absolute bands require a judgement about where to set the floor and the ceiling, so the rationale for each band is published on the metric page, next to the figure.

Two ceilings are taken from the EU Digital Decade targets for 2030. The 80% ceiling for adults with at least basic digital skills is the target for that indicator. The 75% ceiling for enterprises buying cloud services uses the target that 75% of enterprises take up cloud computing, big data or AI. Because that target covers the three technologies together, we do not use it as the ceiling for enterprise use of AI. Every other band is our own judgement, and its rationale is published with the metric.

EU member states with a population below four million, together with Ireland, a small open economy with a large digital sector, and Greece, the closest comparator in language and economic structure. Nine countries in all, including the country being measured.

CY CyprusMT MaltaEE EstoniaLV LatviaLT LithuaniaSI SloveniaLU LuxembourgIE IrelandEL Greece

We publish a peer comparison for a metric only when figures are available for every country in the group. A comparison with gaps would produce a rank that appears more precise than it is. 4 of the 14 metrics have no peer comparison, and their pages say so.

Each figure can be checked independently. Its definition states what is counted and what is excluded, and its source link leads to the document the figure was taken from. Negative findings, such as the absence of a EuroHPC AI Factory in Cyprus or of a qualifying AI model developed there, are the figures most likely to be wrong, because a search of an open register cannot prove that something does not exist. The pages for these metrics name the registers that were checked and the date of the check, so that a counter-example can be tested against the same criteria. When a figure is found to be wrong, we correct it and update its review date.

Please send corrections, counter-examples and questions about a figure to info@socait.com, quoting the metric ID (for example EN-01.1) and the review date shown beside the value. Cyprus AI Monitor’s own figures are published under CC BY 4.0 and may be reused with attribution. Each metric is also subject to the licence of its original source, which is shown on the metric page and in the API.