Baidu Implements Automated AI Optimization for GPU Inference

Baidu's ERNIE research team introduced an automated process called ERNIE Evolve to optimize inference speed on Baidu's Kunlun P800 chips. The team reported that this process achieved 623.6% of baseline speed in a post dated November 9, 2026. However, the same report noted smaller gains of 33.7% and 14.9% when optimizing GPU-based inference, and Baidu stated its GPU baselines were already tuned by its engineers.

The work demonstrates Baidu's efforts to improve the efficiency of its AI operations using automated research loops. This optimization is relevant because Baidu's cloud services rely on the utilization of GPU technology to drive its AI financial growth, putting it in competition with companies like Alibaba Group.