Core Findings
China’s AI-pharma sector has crossed the threshold from concept validation into clinical and commercial verification. Unlike Russia’s sovereignty-driven model or Ukraine’s wartime-adaptive approach, China’s trajectory exhibits three distinctive features: dual propulsion from state industrial planning and private capital, the world’s first Phase III trial of a generative-AI-designed drug, and a bifurcated ecosystem where internet giants and specialized AI biotechs occupy fundamentally different niches. Yet critical bottlenecks persist: no AI-designed drug has yet received regulatory approval, Phase II success rates remain a “Darwinian sea,” and the industry faces tension between a “cost-and-speed” narrative and genuine breakthrough innovation.
- Policy Framework: AI-Pharma Embedded in National Industrial Strategy
In September 2026, ten Chinese ministries and commissions jointly issued the 14th Five-Year Plan for Pharmaceutical Industry Development, explicitly incorporating “computational medicine” as a specialized term into national planning for the first time. The plan mandates accelerating AI, quantum computing, supercomputing, and computational medicine to empower drug discovery . Quantitative targets include: First-in-Class (FIC) drugs accounting for over 25% of global output by 2030, innovative drug industry revenue growing at an average annual rate exceeding 20%, and listed innovative drug companies achieving average R&D intensity above 10% .
The plan designates nine specialized columns for AI-pharma applications, specifying key scenarios including “AI-empowered target screening, drug molecule design and optimization,” and covering the full chain from discovery to clinical research, process optimization, quality management, and smart regulation . Crucially, the framework aims to build a “pharmaceutical and health trusted data space” to facilitate data circulation—addressing a structural bottleneck in AI-driven drug discovery.
Expert interpretation emphasizes that “computational medicine” differs fundamentally from conventional “AI-pharma.” The former integrates machine learning, knowledge graphs, multi-omics integration, causal inference, and digital twins to build “verifiable, reusable, and deployable real disease models,” rather than confining itself to protein structure prediction or molecular screening . This distinction signals that Chinese policymakers recognize algorithmic iteration alone cannot break the “double-ten curse” of drug development.
- Clinical Milestone: The World’s First Generative-AI-Driven Phase III Trial
The most concrete evidence of China’s progress comes from Insilico Medicine (HKEX: 3696). In September 2026, the company announced that Rentosertib (ISM001-055) had completed first-patient dosing in its Phase III trial GENESIS-IPF-3—the world’s first Phase III clinical trial of a generative-AI-driven innovative drug .
Rentosertib targets TNIK, a protein never previously linked to fibrosis. The Phase III trial is a 52-week, prospective, multicenter, randomized, double-blind, placebo-controlled study expected to enroll 320 patients across 47 centers in China, with the primary endpoint being the annual rate of decline in forced vital capacity (FVC) . Leading PI Professor Xu Zuojun of Peking Union Medical College Hospital stated: “TNIK, the target driven by AI, had never previously been linked to fibrosis. This perhaps indicates that AI is carving out a path distinct from traditional research paradigms in target discovery for complex diseases” .
The drug’s early discovery and development was completed in under 18 months, with fewer than 80 molecules synthesized and tested—a fraction of the thousands typically required . In the Phase IIa trial, the highest-dose group showed a mean FVC increase of 98.4 mL, while the placebo group declined by 62.3 mL, yielding a difference exceeding 160 mL . This data was published in Nature Medicine in 2025, with earlier discovery work documented in Nature Biotechnology in 2024 .
Insilico’s commercial model is also notable. As of mid-2026, the company reported signed contract value of approximately $11 billion** since 2021, with **$7.3 billion in new contracts in the first half of 2026 alone. H1 2026 revenue reached $106.3 million, up 287.2% year-on-year, achieving its first profitable half-year since IPO .
- Capital and Ecosystem Divergence: Internet Giants vs. Specialized Biotechs
China’s AI-pharma ecosystem exhibits a pronounced “dual-track” structure.
Internet Giants. ByteDance spun off its AI drug discovery unit into Anew Labs, which raised $290 million** at a **$1.5 billion valuation, with ByteDance retaining a 56% stake. Investors included HSG (formerly Sequoia China), IDG Capital, and 5Y Capital . Anew Labs operates Protenix, an open-source reproduction of AlphaFold 3, and has four drugs in its pipeline including an oral small-molecule immunology candidate .
Baidu-backed BioMap secretly filed for Hong Kong IPO in March 2026. Its xTrimo V4 large model has 268 billion parameters, serving over 800 institutional users and 30+ top enterprise clients, including a partnership with Sanofi . BioMap claims its BioMap OS system has been validated in over 60 proof-of-concept projects, compressing traditional 5–10 year drug discovery cycles to 1–2 years with 70%+ cost reduction .
Specialized AI Biotechs. Jitai Technology (????) listed on HKEX in May 2026 as the “first AI drug delivery stock,” surging 126.67% on debut with a market cap of HK$27.4 billion. Its NanoForge platform includes a lipid library of over 10 million structures, enabling targeted delivery to 8 key organs and tissues . The company has over 10 pipeline products, 3 in clinical stages, and 1 pre-NDA .
- Traditional Chinese Medicine Digitalization: A Distinctive “China Track”
A unique dimension is the application of AI to Traditional Chinese Medicine (TCM). The 2026 Global Digital Economy Conference’s Digital TCM Forum showcased the “Xinhuo Zhongguo Yao” model—China’s first nationally filed TCM discipline large model—along with embodied AI robot “Xiao Zhi” and multiple TCM standards for intelligent agents . Beijing is positioning itself as an “innovation source highland” for digital TCM, supporting a trusted data space for TCM clinical data and advancing clinical-research integration .
However, the core challenge remains: TCM’s “syndrome differentiation and treatment” logic is fundamentally in tension with AI’s requirements for standardization and datafication. No rigorously validated AI-driven TCM new drug has yet emerged.
- Bottlenecks and Forward Assessment
Bottleneck 1: Phase III attrition. Data shows AI-designed molecules achieve 80–90% success in Phase I, significantly higher than the ~52% historical average, but Phase II success rates of 30–40% align with industry norms . Whether Rentosertib replicates its Phase IIa dose-dependent efficacy trend in 320 patients over 52 weeks will determine industry confidence.
Bottleneck 2: The limits of “cost-and-speed” narrative. Some founders argue that “cost reduction and efficiency improvement is a false proposition at this stage,” noting Chinese companies have already compressed high-quality preclinical data package costs to $2–3 million. AI’s true value must lie in “solving previously untreatable diseases and discovering previously non-existent mechanisms” .
Bottleneck 3: Data quality and closed-loop capability. Insilico’s co-CEO states AI-pharma companies must prove not merely “can design molecules” but “can discover targets with novel biological mechanisms” . This requires integrating multi-omics data and performing causal inference, not just structure prediction.
Forward Assessment. China’s AI-pharma sector has achieved algorithmic parity with global leaders, combined with preclinical validation speed 2–3x faster and costs one-half to one-third of Western counterparts. If Rentosertib’s Phase III data are positive and approval follows, China will establish a “first validation loop” advantage in the global AI-pharma race. If Phase III fails, the industry will be forced to re-examine the true efficacy boundaries of AI in complex disease target discovery. TCM digitalization represents a separate long-term variable whose success depends on building verifiable computational models while respecting the logic of syndrome differentiation.
