Can economies innovate their way out of climate trouble? I provide the first systematic evidence outside agriculture that firms adapt to extreme heat through directed technological change. Linking firmlevel data with patent records for nine EU countries (2000–2020), I establish three findings. First, extreme heat operates as a labor-biased productivity shock: labor-intensive firms suffer disproportionate losses and cede market share to capital-intensive competitors. Second, firms respond by shifting production toward capital and redirecting innovation toward labor-saving technologies, with the strongest responses in labor-intensive industries facing the greatest heat exposure. Third, this endogenous innovation response has quantitatively meaningful consequences: labor-saving patents filed in response to heat attenuate aggregate productivity losses by 26 percent over the study period. These findings demonstrate that innovation is not merely a driver of growth but an active margin of climate adaptation.

Using establishment-level data from Orbis during 2000–2020, we study the effect of extreme temperature shocks on local industry market power. We document that extreme temperature exacerbates local market power, increasing industry concentration and aggregate markup. This increase in market power is driven by the heterogeneous responses of firms to temperature-induced productivity shocks across different firm sizes. Small firms are more vulnerable to temperature shocks compared to large firms, leading to a reallocation of market shares from small to large firms under climate change. Compared to large firms, small firms’ productivity is more vulnerable to temperature shocks. This differential response increases market concentration and markups. We rationalize the empirical finding with a model of reallocation and endogenous markup. By calibrating the model to the observed aggregate markup and concentration, we quantify the welfare loss of climate-induced productivity shocks. We find that welfare quantification under standard CES assumption significantly underestimate the welfare loss, as it fails to account for the role of endogenous markup and its consequent misallocation.