For the last 5 decades, the main factors of the world timber demand are the construction dynamics, the demographics, the user's income, the price and availability of wood products relatively to alternative materials, and regulations and policies. We measured the long run demand elasticities the major regions of the world, and built phenomenological models which can predict the wood demand up to 2050 with known e1rnrs. For the prediction of the wood demand until 2050, factors such as GDP and prices which would only rely on arbitrary scenarios, are discarded, and we measure how much of the variability of the wood demand can be predicted from phenomenological and robust variables such as demographic factors. We find that wood demands of the tropical regions (Latin-America, Sub-saharan Africa, and Tropical Asia) can be reliably modelled with good prediction errors bellow± 13% to± 24%. Conversely, the demand models of the temperate regions where the wood dynamics are more linked with the discarded variables, are less precise and their prediction errors escalate up to± 60%. We find that, by 2050, tropical regions will demand between 800 and 1100 million m3 of wood per year, with 95% of confidence. Until now most of the demand by tropical regions was met by a local production, totalling around 500 million m3 per year in 2016. Our findings raises the question of how, by 2050, tropicalforests would be able to sustainably produce the double of what they produce today?