Objective: Cannabis demand (i.e., relative value), assessed cross-sectionally via a hypothetical marijuana purchase task (MPT), has been associated with use, problems, and dependence symptoms, among others. However, limited work exists on the prospective stability of the MPT. Furthermore, cannabis demand among veterans endorsing cannabis use, and the prospective cyclical relationship between demand and use over time, have yet to be investigated. Method: Two waves of data from a veteran sample (N = 133) reporting current (past 6-month) cannabis use were analyzed to assess stability in cannabis demand over 6 months. Autoregressive cross-lagged panel models (CLPMs) assessed the longitudinal associations between demand indices (i.e., intensity, Omax, Pmax, breakpoint) and cannabis use. Results: Baseline cannabis use predicted greater intensity (β = .32, p < .001), Omax (β = .37, p < .001), breakpoint (β = .28, p < .001), and Pmax (β = .21, p = .017) at 6 months. Conversely, baseline intensity (β = .14, p = .028), breakpoint (β = .12, p = .038), and Pmax (β = .12, p = .043), but not Omax, predicted greater use at 6 months. Only intensity demonstrated acceptable prospective reliability. Conclusions: Cannabis demand demonstrated stability over 6 months in CLPM models, varying along with natural changes in cannabis use. Importantly, intensity, Pmax, and breakpoint displayed bidirectional predictive associations with cannabis use, and the prospective pathway from use to demand was consistently stronger. Test–retest reliability ranged from good to poor across indices. Findings highlight the value of assessing cannabis demand longitudinally, particularly among clinical samples, to determine how demand fluctuates in response to experimental manipulation, intervention, and treatment.