Foundation-model providers observe rich usage histories and sell technologically distinct tokens to users running heterogeneous and expanding applications. Users choose tasks involving model- and workflow-specific token recipes predicted imperfectly beforehand, whereas application purpose, prototype success, and downstream revenue remain difficult to contract on directly. I characterize the joint determination of the providers' optimal token pricing, application discovery in the ecosystem, and industrial organization of foundation models, to inform platform design, enterprise procurement, and policy. The same menu governing both current and prospective applications renders discovery dependent on the density of marginally profitable applications and the metered share of development costs. Varying application populations alters optimal billing and price discrimination, which steers discovery. Introductory prices depend on retainable demand, whereas nontransferable compute credits can bridge external financing gaps. Regardless, better token-consumption forecasts encourage experimentation under uncertainty. Finally, competition lowers markups for portable applications while weakening providers' incentives to fund discovery.