China’s AI talent race sends pay soaring, even for interns

The intense competition for AI talent in China reflects a deeper struggle for proprietary knowledge, or "recipes," in model training. Unlike the internet era, where entrepreneurial experience was widespread, the specialized knowledge required for large-scale foundation model training is concentrated in a very small pool of researchers. These individuals hold insights into design choices, data mixes, and training methodologies that are difficult to document or patent, making their movement between companies a direct transfer of critical, hard-won expertise.

This dynamic has led companies to adopt unorthodox recruitment and retention strategies. Beyond soaring salaries and equity incentives, firms are offering authority, university affiliations, and even systematically targeting rivals' key personnel with offers to join or start new ventures. ByteDance, for instance, increased bonuses and adjusted salary caps, while Tencent hired a young OpenAI researcher to overhaul its Hunyuan operation, demonstrating a willingness to disrupt internal structures to attract top talent.

However, the focus of talent demand is shifting. While algorithm specialists were once paramount, their recruitment is slowing as major companies fill core teams. The new priority is for roles that turn models into revenue-generating applications, such as forward-deployed engineers and AI agent product developers. This indicates a market maturation where the emphasis moves from foundational model development to practical deployment and commercialization, potentially altering the long-term career prospects for AI researchers.

The escalating costs of AI infrastructure, with major Chinese tech companies spending billions on computing resources, also suggest that the current, near-limitless spending on researchers may not be sustainable indefinitely. Companies are becoming more selective, narrowing their hiring to candidates with direct experience on the strongest models and teams. This trend points to a future where access to elite academic institutions and frontier projects becomes an even more critical differentiator for aspiring AI professionals.