An MIT News report dated Oct. 6, 2026, described the Lincoln AI Computing Survey (LAICS), a project of the Lincoln Laboratory Supercomputing Center that has summarized commercial AI accelerators and compared their peak performance and peak power since 2018. The first paper in the series studied 57 accelerators, while the latest examined more than 120, according to the report.

The team, led by staff member Albert Reuther and including Michael Jones, Peter Michaleas, Jeremy Kepner and Vijay Gadepally, draws all data from public sources and sorts accelerators by whether they are on a chip, card or system. The report said some companies keep performance and power data private. Each paper also explores a new aspect of the field; the latest examined how architectural choices such as more cores per processor or parallel performance would change a system. The 2022 paper attributed performance gains to smaller, denser transistor designs and lower numerical precision.

Reuther said the survey is planned to continue, and that six new startups had announced first AI accelerators in the preceding few months.