This paper explores the least absolute deviation (LAD) estimator of the autoregressive model with heavy-tailed G-GARCH(1, 1) noise. When the tail index α∈(1,2], it is shown that the LAD estimator asymptotically converges to a linear function of a series of α-stable random vectors with a rate of convergence n1−1/α. The result is significantly different from that of the corresponding least square estimator which is not consistent, and partially solves the problem on the asymptoticity of the LAD estimator when the tail index is less than 2. A simulation study is carried out to assess the performance of the LAD estimator and a real example is given to illustrate this approach.