Our full data collection and preparation methods are described in detail in the documentation accompanying the full dataset (available for download). As in prior years, we prioritized data collected directly from funders and from centrally reported sources of record, and focused on the open infrastructures (OIs) catalogued in IOI's Infra Finder in order to tie the analysis back to the additional attributes those entries provide.
The principal change this year: we used a large language model classifier to screen out false-positive matches and to categorize awards, steps performed by manual review in previous editions. The full documentation reports how each classifier performed against human-annotated data, along with its known limitations.
Briefly:
- Sources. We harvested data directly from funders’ websites when it was available; we also harvested from OpenAIRE, OpenAlex (new this year), 360giving, and usaspending.gov. For several private foundations, we reviewed grants reported on US IRS Form 990 filings. This dashboard covers awards with start dates from 2001-2025 inclusive; awards with no date are included in counts but not in amounts. The downloadable dataset also includes awards from before 2001.
- Selection. We searched award titles, descriptions, and recipients using a predefined list of search terms. A classifier then removed matches that mention an OI only in passing (e.g., as an illustrative example) or that coincide with an OI’s name; uncertain cases and awards with too little text were routed to human review. Based on our testing, roughly one in six false matches is retained.
- Deduplication. We removed awards that appeared more than once, including the same award recorded by different sources under different identifiers. Deduplication at this scale is imperfect, and some duplicates remain..
- Currency. We converted non-USD amounts to USD using European Central Bank historical rates at each award’s start date. Undated awards are counted but carry no USD amount.
- Categories. Direct awards are those where funds reach the OI or its host organization; this dataset and dashboard currently treats any award reaching the host as Direct, regardless of its stated purpose, a rule we are now reviewing [see errata]. Direct awards are further categorized as research and development, operations, and other. Adjacent awards do not flow to an OI but support an activity that uses or builds on one, for example depositing a research work to an open repository. Adoption awards support uptake of an OI by a user or community. Unknown awards are those whose relationship to the OI we could not determine. The classifier under-identifies Adoption and Unknown awards, so treat those counts as lower bounds.
- Multi-recipient awards. We counted multi-recipient awards for each named OI, wherever total counts are reported. Because we rarely know how funds were divided among collaborators, their amounts are not attributed to individual OIs but are included in overall totals.
- Consortial awards. Where an OI is one participant in a large multi-partner consortium, the headline award total overstates its funding. For the largest awards (the top 20 by value, plus European Commission awards over $2M), we checked participant-level records: where an OI's share could be recovered, only that share is counted; where it could not, the award is counted without an amount. Smaller consortial awards were not reviewed and may still overstate funding for some OIs.
This work was generously supported by the sponsors and partners of the State of Open Infrastructure 2026 report: Special thanks go to Crossref, the report’s Supporting Sponsor, and to our Community Supporters: the Research Data Alliance (RDA), the Council on Library and Information Resources (CLIR), the American Geophysical Union (AGU), and Cassyni.
