How it works
The Wikipedia archive functions as a soft sensor for collective attention toward compute infrastructure concepts.
Daily pageview statistics from the Wikimedia pageviews service are used as indirect measurements of when practitioners actively investigate technical topics.
The archive is not intended to measure infrastructure adoption. It measures when users actively look up technical concepts such as artificial intelligence models, cloud platforms, GPUs, data centres and related systems. In this sense the dataset acts as a sensor for conceptual investigation within the compute ecosystem.
Epoch 1 observations
Observations are collected deterministically, deduplicated into canonical records and stored in an append-only dataset. Dataset state hashes are anchored on the Algorand blockchain through the COSTS asset so that the archive can be independently verified. The public anchor can be inspected at https://explorer.perawallet.app/asset/3456128846/
The first observation window, epoch 1, spans 1 January 2025 to 16 February 2026 and contains 8140 daily observations across 20 tracked articles. The analysis collapses the entire period into a robust baseline using median statistics in order to suppress short-lived spikes caused by news cycles or online discussions.
Within this dataset attention directed toward cloud infrastructure concepts and attention directed toward local compute concepts appear in the same order of magnitude. This indicates that local compute topics form a substantial component of the infrastructure discussion within the Wikipedia environment. Because the archive measures conceptual lookup behaviour, the result should be interpreted as one signal within the broader ComputeCosts observatory rather than as a direct indicator of deployment.
Reporting model
To keep the analysis reproducible, the observatory publishes reports in discrete epochs. Each epoch freezes the dataset at a specific moment and produces a deterministic analysis snapshot anchored on the Algorand blockchain.
Epoch 1 is the first baseline of the Wikipedia soft sensor archive.
In parallel the collection agents continue recording pageview statistics, meaning later epochs will appear naturally over time and will form a long term record of how interest in compute infrastructure concepts evolves.
Epoch 1 report
The full methodology, dataset description and analysis code are documented in the epoch 1 report.
Download the report: compute_costs_observatory_report_001.pdf