Despite rapid advances in information technology (IT), corporate investment in new technologies remains slow and uneven across firms. This paper provides new empirical evidence that skill mismatch—the misalignment between firms’ skill requirements and workers’ capabilities—acts as a central friction in the diffusion process. Using large language models to construct granular measures of skill demand and supply from matched data on firm job postings and worker resumes, we test the predictions of a model in which skill mismatch evolves endogenously with firms’ investments in different IT vintages. We document a U-shaped pattern: skill mismatch is highest when new technologies are introduced, declines as firms and workers adjust, and rises again as technologies become obsolete. These gaps are substantial not only for technical skills but also for IT-complementary nontechnical skills such as managerial and organizational capabilities. We further show that firms invest more in intangible capital and exhibit lower measured productivity early in the technology lifecycle when mismatch is greatest. The findings illustrate how skill mismatch plays a critical role in both the timing and the realized gains from corporate investment in new technologies.