AI人才3次大迁徙：百度先行、六小虎分化、腾讯「腾笼换鸟」

TL;DR - China’s AI sector has undergone three talent migrations—from Baidu to model startups, then toward ByteDance and Tencent, and now toward post-training and agent teams. The shifts reflect competition moving from foundation-model pretraining to inference, applications, infrastructure, and high-quality data.

• Demand for pretraining specialists has reportedly fallen 30%–40% from its peak, while hiring for reasoning, post-training, and coding-agent expertise is intensifying.
• Strategic retrenchment at startups such as Baichuan and 01.AI released foundation-model talent, with ByteDance and especially Tencent emerging as major destinations.
• AI infrastructure specialists in GPU scheduling, FPGA compilation, and inference optimization are highly sought because they directly determine compute utilization.
• High-quality data is becoming a central competitive advantage as compute availability stabilizes and model architectures reach a perceived near-term plateau.