ByteDance is separating its longest-cycle venture—a shift from 15-second videos to decade-long drug development—into an independent entity with a $1 billion valuation.

On September 16, Reuters reported that Anew Labs, the AI drug discovery company spun off from ByteDance, completed its first external funding round of approximately $290 million, achieving a post-investment valuation of around $1.5 billion. HSG (HongShan China), IDG Capital, and Hillhouse Investment led the round, with 5Y Capital co-leading. Additional investors included Gaorong Ventures, Primavera Venture Capital, Boyu Capital, strategic investor SBP Group, and state-backed Shanghai Future Industry Fund. Post-funding, ByteDance retains a 56% stake, maintaining control.

Anew Labs originated from ByteDance’s internal AI for Science drug discovery team, which operated for five years. In late 2020, ByteDance began recruiting AI pharma talent, and by 2021, Liu Kai spearheaded the formation of an official team with about 50 core members. In June 2026, ByteDance spun off the team, algorithm platform, and pipeline assets, officially establishing Anew Labs as an independent entity. Headquartered in Shanghai, it also has offices in the U.S. and Singapore.

Why 50 People Can Support a $1.5 Billion Company

Anew Labs’ core team of roughly 50 consists of AI4S algorithm specialists and pharma R&D experts. Since 2021, Liu Kai has led the team’s growth into a multidisciplinary unit covering algorithms, computational chemistry, and drug development. The spin-off transferred the entire team, technology, and pipeline to Anew Labs, which aims to build a comprehensive AI R&D platform covering all major computational stages of drug discovery, rather than a single-point tool. The company has publicly disclosed a platform addressing multiple computational aspects: AnewFold for protein and molecular complex structure prediction, AnewSampling for molecular dynamics simulation, AnewOmni for full-atom molecular modeling and generation, AnewDesign for antibody design and optimization, and AnewMind, which applies large models to scientific reasoning and decision-making in drug R&D. Public research shows Anew’s technology path is expanding from structure prediction to molecular design, dynamic simulation, and scientific reasoning.

A foundational element of this tech stack is Protenix, previously launched by ByteDance’s Seed team. In 2024, Seed open-sourced Protenix for protein and biomolecular complex structure prediction, with models continuously iterated since. For Anew Labs, Protenix’s value lies not only in the model itself but in proving ByteDance’s R&D capabilities in foundational life sciences models, providing a technical base for future molecular design and drug discovery.

Anew Labs has publicly disclosed four drug development directions, with its IL-17 project being the most detailed. IL-17 is a cytokine involved in immune and inflammatory responses, linked to autoimmune diseases like psoriasis, psoriatic arthritis, and ankylosing spondylitis. Inhibiting IL-17 reduces overactive inflammation, making it a key target in autoimmune drug development. In April, at the American Association of Immunologists (AAI) annual meeting, the company first disclosed its preclinical small-molecule drug AN-5162, which it claims can act on three IL-17 forms: IL-17AA, IL-17AF, and IL-17FF. IL-17 is a clinically validated target with existing antibody drugs commercialized by companies like Novartis, but current IL-17 treatments are primarily antibodies, leaving significant room for small-molecule innovation. Small molecules offer advantages like oral administration, yet the structural diversity of IL-17 subtypes makes design more challenging. AN-5162 targets this gap, aiming to cover multiple IL-17 subtypes with one molecule and advance clinical validation. Presently, AN-5162 is still in preclinical stages, with plans to submit an IND application. In other words, Anew Labs has taken the first step from AI models to candidate molecules, but proving safety and efficacy requires passing the critical clinical trial hurdle.

Why ByteDance Spun Off This “Anti-ByteDance” Company

Anew Labs’ business logic stands in stark contrast to ByteDance’s core internet operations. Douyin’s product iteration cycles happen daily, with algorithm recommendations constantly adjusted based on user feedback, while advertising, e-commerce, and content businesses generate rapid data to validate products. Drug development, however, unfolds over years, from target discovery and molecular design to experimental validation, preclinical studies, and clinical trials. The industry often uses “ten years and one billion dollars” to summarize the high investment and long timelines. AI can compress some stages, but animal studies, clinical trials, and regulatory approvals demand time and data validation—unlike an internet product launch, they aren’t “live on release.”

According to Reuters, sources familiar with the matter said AI drug discovery follows different industry logic and management approaches than ByteDance’s core business, so the split better supports long-term growth—meaning ByteDance will no longer manage this company under internet company performance cycles. External fundraising also introduces a new capital structure for Anew Labs. This round’s investors include financial backers like HSG, IDG Capital, Hillhouse, and 5Y Capital, as well as industrial capital like China Biopharmaceutical and the Shanghai Future Industry Fund. Between AI-generated candidate molecules and real drugs lie synthesis, pharmacology, toxicology, clinical development, and commercialization. Industrial capital can help an AI company fill the gap in the latter half of the pipeline. ByteDance hasn’t sold controlling equity; it still holds 56% post-funding, making this more like “incubation with control”: transforming an internal R&D project into a company with independent capital, incentives, and industry resources, while retaining long-term betting power. This model has precedents. In 2021, Google DeepMind spun off its AI pharma unit into Isomorphic Labs, which completed a $2.1 billion Series B in May 2026 to further its AI drug design engine and internal pipeline. Large model companies are increasingly turning AI for Science from internal R&D capability into independently financed, pipeline-building ventures. While Anew Labs is still far smaller than Isomorphic, it’s on the same trajectory.

The Real Challenge for Anew Labs: Turning Model Capability into Drug Assets

Over the past few years, AI pharma companies’ business models have evolved. Early firms offered tools like protein structure prediction, molecular generation, virtual screening, and drug property prediction, earning revenue through software licensing, technical services, or pharma partnerships. As model capabilities deepen into drug R&D, a wave of companies is shifting from “providing R&D tools” to directly participating in or leading pipeline development. Anew Labs has chosen the latter path, which means its technical evaluation metrics can’t stop at model performance. How accurate structure prediction is or how efficient molecular generation becomes must ultimately answer a stricter question: Can candidate molecules from models pass experimental validation, enter real drug development processes, and become clinically valuable drug assets?

AnewSampling’s molecular dynamics research exemplifies this. Drug discovery deals not with static protein structures but with dynamic systems where proteins, ligands, and their environment constantly change, and molecular binding is inherently dynamic. Modeling these processes requires significant computation and high-quality experimental data. Unlike internet software, where data can be scaled by expanding training sets, much critical drug R&D data must come from wet-lab experiments. Consequently, AI pharma R&D efficiency hinges on establishing a continuous data loop between models and experiments: AI designs molecules, experiments validate them, data feeds back, models iterate, and the cycle repeats. This is the capability Anew Labs must build after spinning off from ByteDance’s internal R&D team. The models, algorithms, and computational infrastructure ByteDance accumulated provide a starting point, but if the ultimate business model revolves around drug assets, experimental systems, pipeline advancement, and clinical development will equally define long-term value.

In the Chinese market, Anew Labs already faces competitors that go beyond mere model capability. InSilico Medicine follows a path of “AI discovery + proprietary pipeline + out-licensing.” On September 10, its core pipeline Rentosertib entered Phase III trials, showing AI-generated drugs advancing to higher-stage clinical validation. The company also commercializes through pipeline licensing and joint development, linking AI R&D capability with drug asset monetization. XtalPi takes a different route, entering drug R&D with AI, computational, and automated experimental platforms, offering R&D services to pharma companies while participating in pipeline collaborations. In 2025, XtalPi generated RMB 803 million in revenue, growing 201.2% year-over-year, and achieved annual profitability. For such companies, commercialization success isn’t solely tied to whether proprietary drugs reach the clinic; it depends on whether platforms and R&D services can consistently create value for pharma partners. BioMap, meanwhile, starts from foundational life sciences models, combining model capabilities with biological experiments, antibody discovery, and drug development, steadily extending toward industrialization. These three companies have formed distinct paths: some convert AI capability into proprietary pipelines, others earn via technical services and collaborations, and some try to connect foundational models all the way to experiments and drug R&D. Anew Labs is entering a market where business models are already being validated. From foundational models like Protenix to candidate drugs like AN-5162, Anew Labs has completed the first leg from algorithmic research to drug discovery. Now it must navigate the industry’s strictest and longest validation: whether experimental data supports model predictions, whether candidate molecules advance into the clinic, and whether these pipelines eventually yield commercially viable drugs. As AI pharma moves from “can it discover molecules” to “can it consistently produce drugs,” a company’s value will ultimately be determined by how many validated drug assets it can sustainably generate.