We study the nonlinear limits to arbitrage in a model. When mispricing is small, arbitrage activity increases with mispricing because of the higher cost-adjusted return. However, at high levels of mispricing, arbitrageurs are deterred by larger mispricing as funding constraints become more binding. Testing the model predictions on the index spot-futures arbitrage with a Markov-switching model, we document an inverse U-shaped relationship between mispricing and arbitrage activity. The extreme regime is with the largest mispricing but least arbitrage activity, and coincides with the market turmoil, suggesting that funding constraints become the main driver behind the limit to arbitrage.