In this index, a chip citation is a paper’s mention of using a particular chip or chip family. It is an indicator of hardware use in published AI research, not a count of citations that the paper itself receives. The source analysis is supplied by Zeta Alpha. It does not measure how many GPUs a lab owns, how many chip units were sold, or how many GPU-hours the research consumed.
How to read this finding
A paper can use more than one chip type, so chip categories can overlap and should not be treated as mutually exclusive shares of all papers. Publication and reporting practices also differ from real-time hardware deployment. The separate research-topic analysis covers 6,356 papers from January 1 to June 1, 2025, uses model-assigned topic labels, and retains chip-topic pairs with at least three papers. Its percentage-point differences describe specialization relative to the sample, not chip speed or market share.
Citation estimates use inputs through September 1, 2026; fleet counts have an October 1, 2026 cutoff. The research-topic sample covers January 1 to June 1, 2025. See each answer for its period and limitations.
Benaich, Nathan. “What does a chip citation in an AI paper measure?” State of AI Report Compute Index. Web page updated 2026-10-11; data periods as specified above.
