Adoption of AI technologies, capability development, and performance inequity among startups in an upper-middle-income country

The analysis investigates the AI adoption behavior, the development of capabilities, and the differences in performance between 302 active venture founders affiliated with Mexico’s Instituto Politécnico Nacional (IPN), a theoretically strategic sample similar to the few that can be used because the high technical literacy setting permits isolating cognitive and normative adoption barriers from the digital access constraints. The integrated framework, which combines the TOE model, the Technology Acceptance Model (TAM), Institutional Theory, the Resource-Based View, Self-Efficacy Theory, and the Productivity J-Curve, posits that AI Implementation Self-Efficacy (AISE) is the key mediating factor between organizational resources and adoption outcomes. Data from the research were collected through an online questionnaire with a satisfaction coefficient of 0.84 and analyzed using descriptive statistics, Mann-Whitney U and Kruskal-Wallis tests, k-means clustering (silhouette coefficient = 0.64), and ordinal logistic regression. Statistics reveal that almost 90% of company founders have a good understanding of AI technologies, with 89% implementing conversational AI tools such as chatbots, while only 16% are involved with platforms that use predictive analytics. Literature reports that trust in the results of AI (M = 3.28) and ethical concerns (M = 3.15) have been considered the main barriers to adoptions (they have a higher impact than the costs or required technical skills), with women voicing the concerns more strongly than men (p < 0.05). J-curve performance was shown to be part of this experience, with the short-term product and the long-term product, as well as sales benefits, following the innovation and productivity effects. The three types of adopters were: Confident Adopters (42%), Pragmatic Experimenters (38%), and Barrier-Constrained (20%), with short-term gains and long-term sustainability benefits, across the Technology/Health and Trade/Agribusiness sectors. These results indicate that in an institutional environment of ambiguity, AI adoption is primarily a cognitive and normative issue rather than a resource issue, and that it has consequences for trust-building policies, sector-specific training, and future longitudinal research.