By | Zhou Xinyu
Edited by | Zhang Yuxin
Intelligencer exclusively learned that ByteDance's AI drug discovery business line has initiated the process of spinoff and independent financing.
It is reported that ByteDance will still hold a controlling stake in the new company post-spinoff. The core AI drug discovery team, core algorithms, technology platforms, and existing pipeline assets will be transferred in their entirety to the new entity. Meanwhile, this business will continue to receive computing power support from Volcano Engine.
The new company will be led and managed by ByteDance's AI drug discovery team. The ByteDance AI drug discovery team was established in 2021, headed by Liu Kai. It is understood that the core team consists of approximately 50 members, comprising AI4S algorithm talents and seasoned pharmaceutical experts. Since its inception, this team has undertaken the core function of transitioning from fundamental model research to industrialization.
Previously, the internal team responsible for protein structure prediction models at ByteDance has also been merged into the AI drug discovery team led by Liu Kai. The relevant algorithm model teams have completed integration and will continue to advance fundamental model research in this field, with a few personnel having left.
The business progress of ByteDance's AI drug discovery serves as a crucial foundation for this spinoff and financing round.
ByteDance has achieved multiple technological results in the direction of AI drug discovery. In 2025, the ByteDance AI4S team released the molecular structure prediction models Protenix and Seedfold, and iterated on Protenix-v1/v2 in 2026, building high-precision open-source structure prediction capabilities for biological complex systems such as proteins and ligands.
In protein design and prediction, the team has launched tools like PXDesign for designing protein binders.
Simultaneously, ByteDance also introduced the AI drug discovery platform Anew Labs, targeting real-world drug R&D.
As shown on the Anew Labs website, the team has released research including AnewSampling, AnewOmni, AnewFEP, AnewSynth, scNext, covering areas such as protein-ligand dynamic structure prediction, all-atom molecule generation, free energy calculation, synthetic feasibility prediction, and virtual cells. It has also unveiled early-stage drug pipelines like IL17AA/AF/FF and IL4R.
In April 2026, Anew Labs first disclosed the IL-17 small molecule project at the American Association of Immunologists Annual Meeting, achieving, for the first time globally, the blockade of three IL-17 family dimers (AA/AF/FF) using small molecules. Since IL-17 is a key pathway in autoimmune diseases like psoriasis and ankylosing spondylitis, and simultaneously inhibiting A/F (two key inflammatory factors) has been clinically validated as valuable by antibody drugs.
This indicates that ByteDance's AI drug discovery capabilities have progressed beyond models and algorithms, further entering the stage of specific target validation, specific molecules, and internal pipeline verification.
With technological advancements and continuous progress in AI drug discovery exploration, ByteDance judges that the opportunity from scientific research to industry is maturing. Therefore, it has decided to integrate various internal teams and test the waters of industrialization.
Of course, the industrialization of AI4S poses significant challenges.
The validation cycle for AI4S businesses is long, and the process is more complex. Taking drug discovery as an example, it encompasses multiple complex stages such as model development, wet lab experiments, and clinical validation. Consequently, it demands a larger pool of professional talent and requires organizational and management forms that differ from those of internet businesses.
Informed sources revealed that this business spinoff is aimed at establishing an independent organizational structure more suited to the characteristics of this business. ByteDance hopes that through this adjustment, it can better attract top-tier talent to join, thereby advancing fundamental model capabilities in this field and the integration of algorithms with the pharmaceutical industry.
Meanwhile, the pharmaceutical industry itself is under efficiency pressure.
Over the past two decades, global pharmaceutical R&D investment has continuously increased. IQVIA, one of the world's largest providers of advanced analytics, technology solutions, and clinical research services to the life sciences industry, estimates that global medicine spending will reach approximately $2.3 trillion by 2028.
While the market size is substantial, the core pain points of high R&D costs, long cycles, and high failure rates in new drug development have not fundamentally changed. The industry urgently hopes to introduce AI technology to break through these limitations.
Currently, AI4S research is accelerating, reflected in its significantly enhanced ability to solve complex problems.
Taking the iteration of the AlphaFold series (protein structure prediction models developed by Google DeepMind) as an example: from the initial version proving feasibility, to AlphaFold 2 achieving atomic-level accuracy predictions for 200 million proteins, to AlphaFold 3 breaking the limitation of single proteins and accurately predicting complex interactive systems — this demonstrates that AI has deeply penetrated critical aspects of drug design.
If protein structure prediction was still a fundamental research problem, then the emergence of multimodal molecular generation models in recent years directly addresses the core issue of the pharmaceutical industry — drug design. This may also indicate that AI drug discovery is gradually moving from research towards industrial application.
ByteDance has been involved in AI4S for many years. As early as around 2020, ByteDance began systematically entering fields like AI drug discovery, molecular simulation, and computational biology. Subsequently, it had teams covering directions such as first-principles calculations, quantum chemistry, molecular dynamics, materials simulation, as well as energy and drug molecule generation.
After the establishment of the Seed team focused on large model research, AI4S also became part of ByteDance's frontier technology layout.
A person close to this spinoff stated that this is ByteDance's first attempt at the industrialization of AI4S, and it is taken very seriously internally. "Biotechnology has its own industrial logic. Spinning it out independently provides decision-making flexibility, with the hope of paving a viable industrial path for AI4S in China."
