PILLAR 2 OF 7

Compute

EuroHPC AI Factories and AI Factory Antennas hosted locally, and the number of data-centre-class GPUs available through open-access calls at the national HPC facility.

pillar score

30/100

3 metrics · 3 tier A · 0 live

Show the exact score calculation30/100

How this pillar score is derived

Each value is converted to a metric score using its published reference band. The metric scores are then combined according to their relative weights.

Absolute reference bands · version 1.0.0 · in force from 7 Aug 2026

EuroHPC AI Factories hosted

CO-01.0 · Higher is better

0.0/100

Raw value: 0
Reference band: 0 → 1
Relative weight: 3.0/8.0
Pillar contribution: 0.0 / 100

Worked equation

100 × (0 − 0) ÷ (1 − 0) = 0.00

European High Performance Computing Joint Undertaking · 2025 round 3 · 10 Oct 2025

Why this reference band?

The metric asks whether Cyprus hosts an AI Factory at all. Hosting one is the difference between having a supercomputer optimised for AI in Cyprus and reaching one across a border. We set the ceiling at one because a second factory would change much less than the first.

EuroHPC AI Factory Antennas

CO-02.0 · Higher is better

100.0/100

Raw value: 1
Reference band: 0 → 1
Relative weight: 2.0/8.0
Pillar contribution: 25.0 / 100

Worked equation

100 × (1 − 0) ÷ (1 − 0) = 100.00

European High Performance Computing Joint Undertaking · 2025 selection · 13 Oct 2025

Why this reference band?

One Antenna gives users in Cyprus a route to supercomputing optimised for AI. A second would add another way into the same kind of access rather than a new capability, so we set the ceiling at one.

Open-access GPUs

CO-03.0 · Higher is better

12.5/100

Raw value: 64 GPU
Reference band: 0 GPU → 512 GPU
Relative weight: 3.0/8.0
Pillar contribution: 4.7 / 100

Worked equation

100 × (64 − 0) ÷ (512 − 0) = 12.50

The Cyprus Institute · 2026 register · 7 Aug 2026

Why this reference band?

We set the ceiling at 512 GPUs, the scale at which a facility such as The Cyprus Institute’s could pretrain a small foundation model on its own hardware instead of only fine-tuning an existing one. We chose a figure that should stay meaningful across hardware generations.

Final weighted average

round((0.0000 × 3.0 + 100.0000 × 2.0 + 12.5000 × 3.0) ÷ 8.0) = 30

Exact result 29.69; published score 30/100. The calculation uses unrounded metric scores. The final weighted average is rounded to the nearest whole number.

Read the full methodology

Metrics in this pillar

value first, then score

Sources: European High Performance Computing Joint Undertaking, The Cyprus InstituteDownload data

EuroHPC AI Factories hosted over timeCO-01.0

count, irregular. The vertical axis does not start at zero.

Source: EuroHPC JU AI FactoriesUpdated 10 Oct 2025

GAP TO PEER MEDIAN

EuroHPC AI Factories hosted+0EuroHPC AI Factories hosted: +0 compared with the peer median
EuroHPC AI Factory Antennas+1EuroHPC AI Factory Antennas: +1 compared with the peer median

belowpeer median = 0above

Each bar shows the difference as a share of the metric’s reference band, so metrics in different units can be compared. The labels give the difference in each metric’s own unit.