Dice Question Streamline Icon: https://streamlinehq.com

Pre-2019 AI supercomputer outliers: coverage gap or genuine size

Determine whether the large AI supercomputers identified in 2017–2018 were genuinely the largest deployed systems until 2021 or whether their apparent size above the post‑2018 trend arises from incomplete coverage in the compiled 2019–2025 dataset of AI supercomputers.

Information Square Streamline Icon: https://streamlinehq.com

Background

The paper reports that several AI supercomputers from 2017–2018 appear significantly above the performance trend observed after 2018. Because the dataset’s systematic collection begins in 2019 and coverage prior to that is limited, it is unclear whether these early systems reflect genuine deployments or artifacts of incomplete data collection. Clarifying this would improve historical trend analysis and interpretation of early AI supercomputer deployments.

References

We found several large AI supercomputers in 2017 and 2018, significantly above the trend suggested by our post-2018 results. It is unclear to what extent this reflects a lack of coverage in our dataset or whether these genuinely were the largest deployed systems until 2021.

— Trends in AI Supercomputers (2504.16026 - Pilz et al., 22 Apr 2025) in Section 3.1 (Results: Computational performance of the leading AI supercomputers has doubled every nine months)