We develop a dynamic model of Defi lending incorporating the following key features: 1) borrowing and lending are decentralized and anonymous where terms are set by smart contracts; 2) lending is collateralized on the market value of crypto assets; 3) lenders supply assets to a liquidity pool and indifferent to collaterals pledged by borrowers conditional on the terms in smart contracts. The underlying friction is the limited commitment and asymmetric information between borrowers and lenders, making haircut a key parameter trading-off risk and efficiency. We identify a price-liquidity feedback loop in Defi lending: the market outcome in any given period depends on agents’ expectations about lending activities in future periods, higher future price expectation leading to more lending and higher price today, leading to multiple self-fulfilling equilibria. Defi lending makes crypto prices more sensitive to fundamental shocks, and asset prices and lending activities can fluctuate according to non-fundamental market sentiment. Flexible updates of smart contract terms can restore equilibrium uniqueness. We also discuss some evidence.