Gu Quanquan officially bids farewell to ByteSeed; the next stop may be AI-powered drug development.

June 2nd, Gu Quanquan, a core member of ByteDance's Seed team focused on Scientific Intelligence (AI4S), announced his resignation, ending his three-year research career at ByteDance. In his farewell letter, Gu Quanquan confirmed that the day was his last at ByteDance and emphasized that he would continue to explore in the direction of scaling in the future. Market sources pointed out that his next stop may involve entrepreneurship in the fields of AI-driven drug discovery and protein design, and he has already received follow-up interest from leading US dollar funds. Gu Quanquan’s departure and his potential entrepreneurial moves reflect the organizational boundaries of big tech companies while exploring cutting-edge technology and also illustrate the changing logic of talent flows in the current AI industry. Over the past three years, Gu Quanquan played a critical technical leadership role in ByteDance's Seed team. His résumé shows that he is a versatile talent with both a strong academic background and practical engineering capability. He earned his bachelor’s and master’s degrees in Automation and Control Science & Engineering at Tsinghua University in 2007 and 2010, respectively, and his PhD in Computer Science at the University of Illinois at Urbana-Champaign in 2014. Previously, he served as a tenured Associate Professor of Computer Science at UCLA. After joining ByteDance Seed in 2023, Gu Quanquan primarily led AI-driven drug research and development. The core achievements of his team include the protein structure prediction model SeedFold, the protein binder design model SeedProteo, and the DPLM series of protein language models. By early 2025, his business scope expanded to underlying infrastructure for general large models, and he set up a special team within Seed to tackle challenges in scalable training of ultra-large parameter models. Gu Quanquan’s business trajectory clearly demonstrates the development path of AI4S within large tech companies: leveraging computation and algorithms, rapidly building basic infrastructure in vertical fields. Recently, ByteDance Seed's AI4S team underwent substantial organizational restructuring. Previously, there were rumors that the team might split, but sources close to ByteDance confirmed that the team has not been split; instead, management has been transferred to ByteDance’s Vice President of Technology Yang Zhenyuan for unified coordination. This shows that ByteDance still maintains long-term strategic investment in cutting-edge AI research, though business advancement and organizational structure have become more focused. Such adjustment, along with the departure of core personnel, is essentially a result of AI-driven drug discovery’s shift from "proof of concept" to "pipeline advancement." In big tech labs, developing open-source models like SeedFold is their advantage; but for true commercialization of AI drug discovery, challenges like target discovery, building wet-lab experiment systems, and advancing clinical pipelines must be tackled — real and significant practical problems. This not only exceeds the pure algorithm team's business scope, but the long R&D cycles and high risks deviate from the high-frequency, rapid iteration, and standardized ROI commercial models that big tech companies have pursued for years. Therefore, when technological breakthroughs reach a critical point, key talents choose to depart from big tech systems, leveraging accumulated technology and capital to enter more flexible vertical entrepreneurial tracks — a highly rational business decision. Taking Gu Quanquan’s departure as a case in point, the current AI industry's talent flows are undergoing a marked paradigm shift: talent is moving from centralized concentration in computing giants to vertical industry subdivisions. In the early explosion of large models, the industry’s core demand was "concentration of computation and data," with top AI talent gathering in big tech firms. But as underlying technologies mature, the focus is shifting to vertical tracks like embodied intelligence and scientific intelligence, which require strong industry know-how, and talent is seeking real scenarios where technology can be applied. On the other hand, we see a shift from algorithmic platforms to industry entrepreneurship. Early AI scientists at big tech firms often worked on technical platforms, but AI4S is highly non-consensus, and its industrial complexity makes it difficult for scientists to obtain sufficient long-term error tolerance within existing internet KPI frameworks. Thus, key technical staff begin to leave the big tech environment, facing real pain points in physical industries as entrepreneurs. There is also capital’s push behind this trend. In primary markets, capital is increasingly valuing the commercial conversion rate of AI technology in physical industries. Compared to the bottomless investment in general large models, venture capital is shifting its bets to niche fields like AI-driven drug discovery, which feature high industry barriers and clear commercialization milestones. After the infrastructure construction of general large models has come to a phase, the battle for AI to reshape the physical world’s industries is just beginning. Big tech companies seek organizational boundaries in consolidation, while technical experts entering entrepreneurship must complete the final closure of technology within the harsh realities of industries. Risk Warning and Disclaimer The market contains risks, and investment should be approached cautiously. This article does not constitute personal investment advice and does not take into account individual users’ particular investment aims, financial status, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their specific situation. Any investments made accordingly are at their own risk.