Usage of graph neural networks will increase for a wide variety of applications likely to violate the assumptions of current deep learning hardware.

Judge at evaluation time; there is no stated horizon. The claim has two parts: (a) use of graph neural networks increased across a wide variety of applications after 2019; (b) these workloads tend to break the assumptions of current deep-learning hardware, which is built for dense tensor operations, because they rely on sparse, irregular operations such as gather and scatter. The claim is hedged ('likely'). True: (a) publication and deployment evidence shows GNN use growing across many domains (for example chemistry and drug discovery, recommendation, traffic, physics, social networks), and (b) hardware or systems literature reports that GNN workloads are poorly served by dense-oriented accelerators. Mostly true: both hold, but growth was narrower than 'a wide variety' or the hardware mismatch was only modest. Mixed: one part clearly holds and the other does not. Mostly false: GNN use grew only marginally or only in a few niches. False: GNN use did not grow, or declined.

Set before the research began.

1. E1 Unlocking the Power of Graph Neural Networks—A Systematic Literature Review of Application-Oriented GNN Studies Computer Modeling in Engineering & Sciences (Tech Science Press) · 27 July 2026

   “Figure 9: Domain-level study categorization from 2019 to 2026. The distribution shows strong representation in usercentric applications, generic benchmark settings, health care, transportation, and finance.”

2. E2 Unlocking the Power of Graph Neural Networks—A Systematic Literature Review of Application-Oriented GNN Studies Computer Modeling in Engineering & Sciences (Tech Science Press) · 27 July 2026

   “Many papers were benchmark-driven and did not map well to a single real-world application domain.”

3. E3 Frontiers | Knowledge mapping of graph neural networks for drug discovery: a bibliometric and visualized analysis Frontiers in Pharmacology · 10 May 2024

   “Since 2019, the number of papers published has been increasing, reaching 243 in 2023.”

4. E4 Architectural Implications of Graph Neural Networks IEEE Computer Architecture Letters (arXiv preprint 2009.00804) · 2020

   “We report both the dense and sparse performance on the CPU/GPU, and report only the dense performance of TPU because it does not support the sparse GEMM.”

E1 is a 2026 systematic review that categorises application-oriented GNN studies from 2019 to 2026. It finds strong representation in user-centric applications, health care, transportation and finance, which supports use across a wide variety of domains. E3 records GNN publications in drug discovery increasing since 2019 and reaching 243 in 2023. E2 adds that many GNN papers were benchmark-driven and did not map well to a single real-world application domain. For the hardware part, E4, a study of the architectural implications of GNNs, reports that the TPU does not support sparse GEMM, so only its dense performance was reported. This shows a mismatch with dense-oriented accelerators, but the evidence on that point is narrow, so the verdict is mostly true rather than fully true.