Plateau and Reorientation: A Comparative Trajectory Analysis of Iranian Artificial Intelligence Research, 2010–2025

Purpose. Iran's artificial intelligence (AI) research expanded rapidly during the 2010s, but existing studies describe a single national output curve without benchmarking it against comparable research systems or testing whether post-2018 changes are distinguishable from those of peers. This study positions Iran within a panel of 29 emerging research systems and examines trajectory, international collaboration, citation visibility, and thematic profile. Design and methods. We assembled 13,986 Iran-affiliated AI articles and country-year indicators for 29 systems (1.2 million works worldwide, 2010–2025) from OpenAlex. Inference combines exact rank-permutation tests, difference-in-differences with Fisher randomisation inference, synthetic control with placebo-in-space tests, paper-level regressions with year and topic fixed effects, and Jensen–Shannon divergence and Balassa-type specialisation indices for thematic structure. Findings. Iran ranked second among the 29 systems by share of world AI output in 2015 and 2021, but sixth in 2025. Its 2018–2025 growth in world share was the lowest of all 29 under the primary corpus definition (−5.7% per year versus a median of +5.9%; exact one-sided p = 0.034) but only the fifth lowest (p ≈ 0.17) when preprints and other work types are included; the slowdown began around 2015, before the 2018 reinstatement of US sanctions, and level-shift estimators (randomisation-inference DiD, synthetic control) were not statistically distinguishable from placebo and were sensitive to the donor pool. The internationally co-authored share rose from 15% to 32% but remains about 9 percentage points below the peer median; China became Iran's most frequent partner after 2018 (0.4% to 5.8% of papers) without displacing Western partners. Papers with foreign co-authors were 15–23 percentage points more likely to be top-10% cited after adjusting for year, topic, and team size, with an exploratory indication that the Western premium diminished after 2018. Iran's thematic profile converged toward the world's (JSD 0.26 to 0.15) but remains under-represented in trustworthy-AI topics (explainable AI index 0.43; adversarial robustness 0.60). Originality. The study replaces single-country description with a comparative, inference-aware design, reports negative and fragile results transparently, and shows that the timing of Iran's relative plateau does not fit a simple sanction-shock narrative.