Trees are long‐lived organisms, exhibiting temporally complex growth arising from strong climatic “memory.” But conditions are becoming increasingly arid in the western USA. Using a century‐long tree‐ring network, we find altered climate memory across the entire range of a widespread western US conifer: growth is supported by precipitation falling further into the past (+15 months), while increasingly impacted by more recent temperature conditions (−8 months). Tree‐ring datasets can be biased, so we confirm altered climate memory in a second, ecologically‐sampled tree‐ring network. Predicted drought responses show trees may have also become more sensitive to repeat drought. Finally, plots near sites with relatively longer precipitation memory and shorter temperature memory had significantly lower recent mortality rates (R2 = 0.61). We argue that increased drought frequency has altered climate memory, demonstrate how non‐stationarity may arise from failure to account for memory, and suggest memory length may be predictive of future tree mortality. Tree growth is driven by precipitation and temperature conditions years prior, a process referred to as “memory.” Over the past century, memory of precipitation in Pinon edulis has become longer and temperature memory shorter, probably due to enhanced drought stress. We show memory shifts may be associated with reduced recent mortality risk, and that models ignoring memory may be more likely to find apparent non‐stationarity.