Delaying a financial transaction can be costly, but the cost of delay is difficult to estimate in traditional finance. I exploit the unique data offering and market design of the Ethereum blockchain to estimate delay costs in decentralized finance (DeFi). Structurally estimating an auction model for different users and transaction types, I find that the average cost of delaying a transaction by 30 seconds is 4.53 US dollars, but the distribution of delay costs is highly skewed to the right. Delay costs are lower for payment transfers and higher for token swaps, latency arbitrage transactions, and high-value transactions. I estimate that the daily welfare loss due to network delay on Ethereum was 7.22 million US dollars, and I apply the delay cost estimates to evaluate the welfare losses under several alternative transaction fee mechanisms.