Sequence recommendation has achieved excellent results in recent years, with an ability to capture changes in user interests over time. With the introduction of the self-attention mechanism, the sequence recommendation model can better capture changes in user interests and obtain better user embedding representations. Many studies have combined sequence recommendation models with contrastive learning, enabling the sequence model to capture more information and improve the expression quality of user embeddings. However, problems associated with existing contrastive learning techniques for sequence recommendation still remain. Multiple forward calculations bring an additional training burden. Meanwhile, performing data enhancement of the input sequence destroys information in the dataset. To solve these problems, we propose a contrastive learning model, ESRec, based on enhancing embedding space data. Data enhancement in the embedding space can avoid multiple forwardta calculations and reduce the burden of model training. Ultimately, two types of random data enhancement are used, simulating the impact of noise in the dataset and improving the robustness of the model. Extensive experimental results on four public datasets illustrate that our model outperforms all baseline models with significantly reduced training time.