Stan uses Markov chain Monte Carlo for inference. MCMC algorithms can do very well at approximating the high-dimensional integrals required to calculate the posterior; estimating 1000-dimensional distributions with MCMC methods is not unusual.