ByteDance Scholarship Goes Global: Tracking the Careers of 67 Winners Over Five Years

AI research is notoriously expensive, and many universities lack the resources to sustain long‑term projects. Notable voices such as Fields Medalist Terence Tao and Stanford HAI director Russell Wald have highlighted the systemic shortage of funding for cutting‑edge models.

Since its launch in 2021, the ByteDance Scholarship has been a low‑profile but growing initiative. In 2026 the program opens to global applicants for the first time, removing regional restrictions and increasing the number of awards from the initial 9 to 20 + slots (exact 2026 figure not yet disclosed). Each student receives ¥200 000, and their mentor receives an additional ¥100 000 research grant—the only scholarship in China that provides a dedicated mentor fund.

The eligibility criteria have also broadened: applications are no longer judged solely on mature, published work; promising intermediate results are now accepted. Last year the scholarship attracted over 500 candidates from 66 universities; the global rollout is expected to draw even more top talent.

Over five years, 67 young researchers have been supported. Winners have come from 23 + universities, led by Tsinghua (23 awardees) and Peking University (15). While most are PhD students, two standout masters—Ren Yi and Huang Rongjie—demonstrated that strong research can compete with higher‑degree peers.

Ren Yi, the first author of the FastSpeech series, published more than 30 papers (10 as first author) with over a thousand citations, later joining ByteDance as a speech researcher and now leading a video‑model team at HeyGen. Huang Rongjie focused on multimodal generation, contributed nearly 20 A‑class conference papers, and his open‑source projects have earned over 10 k GitHub stars before joining Meta FAIR.

The career trajectories of awardees illustrate diverse outcomes: 2023 winner Lu Cheng joined OpenAI and contributed to GPT‑4o image generation and Sora 2; Zhu Qihau became a DeepSeek lead developer; Li Yunfei built a full robotics stack and now researches reinforcement learning at ByteDance; Yu Kaichao co‑founded AI infrastructure startup Inferact, raising $150 M seed funding; Bao Fan co‑founded a video‑generation startup Vidu; Sun Tianxiang, a core developer of the MOSS dialogue model, founded Analemma and secured multi‑million‑dollar angel funding; many others remain in academia as professors or research scientists.

Analyzing the research topics reveals that scholarship winners often work on directions two to three years ahead of mainstream hype. Early work on diffusion model acceleration (e.g., FastSpeech, DPM‑Solver) predated the explosion of Stable Diffusion. Later winners explored embodied intelligence, world models, multimodal large models, and AI‑for‑Science, fields that have only recently become commercial focal points.

ByteDance’s internal research aligns with this long‑term view: since 2020 the company has invested in scientific computing, open‑sourced the protein‑structure predictor Protenix, achieved >55 % MFU on the MegaScale training system (1.3 × the performance of leading open‑source frameworks), and continued to push frontier models through its Seed program.

Overall, the scholarship’s shift to accept early‑stage results reflects a strategy to nurture foundational innovations that may not yet be in the spotlight but have the potential to drive future AI breakthroughs.