Abstract: In the digital economy, firms increasingly exploit algorithms to extract users’ latent commercial value from accumulated data. Rising risks of privacy intrusion, however, induce privacy-sensitive users to opt out of data provision, generating a “digital hermit” phenomenon that affects firms’ data-pricing decisions and social welfare. To capture the strategic interaction between users’ multi-period upload behavior and firms’ dynamic pricing, we develop a Markov game that embeds user privacy heterogeneity, upload decisions, and firms’ algorithmic inference capabilities.The model characterizes the joint dynamics of optimal pricing, upload rates, and algorithmic accuracy. We show that improvements in inference capability raise data value but also heighten privacy anxiety, thereby depressing data supply—a double-edged “algorithmic progress–privacy concern” effect. Moderating firms’ inference capability can increase user participation and support the sustainable advancement of algorithmic technologies. Firms set prices above the social optimum, and social welfare evolves non-monotonically. Optimal dynamic pricing features high compensation during early learning to accelerate data accumulation and technological updating, followed by low compensation in maturity to stabilize participation and profits, yielding a “learning-driven–privacy-constrained” equilibrium path. Numerical simulations highlight the critical roles of the discount factor and learning rate in shaping convergence dynamics.

Key words: Digital hermit, User data, Algorithmic inference capability, Dynamic pricing, Markov game