After the 15th Five-Year Plan, how should AI drug discovery be priced: speed, depth, and modality options

On September 18, the Ministry of Industry and Information Technology and nine other departments issued the "15th Five-Year Plan for the Development of the Pharmaceutical Industry." Artificial intelligence, quantum computing, precise molecular delivery, cell programming, gene editing, and other technologies were simultaneously written into the policy framework, and the market responded immediately: XtalPi (02228) closed up 16.32% that day at HK$8.02, with turnover of HK$1.251 billion; INSILICO rose 13.12%, and Jitai Technology rose 12.04%. XtalPi ultimately led the three AI drug discovery companies, reflecting that the focus of capital has expanded from a single algorithm to the system integration capabilities of computation, experimentation, delivery, and frontier modalities. Policy acts as a catalyst, while industry trends determine the valuation ceiling. Today, the core question in measuring AI drug discovery companies has become: who can continuously produce differentiated assets, whose technology can be reused across projects, and who can select the drug modality with a higher probability of success across different targets. The value of clinical leadership must be confirmed by subsequent data First-generation AI drug discovery companies have already gone through a round of clinical elimination and industry consolidation. When Recursion acquired Exscientia, it disclosed more than 10 clinical and preclinical projects; several months later, the company compressed its priority portfolio to more than 5, paused or abandoned 3 clinical pipelines and 1 preclinical pipeline, among which REC-2282 and REC-994 were both terminated due to insufficient clinical data. This history presents the full meaning of "move fast, fail fast": entering clinical trials quickly can allow early falsification, but it can also carry problems in target selection, patient stratification, and molecule quality into a more costly stage. After this round of reshuffling, the three listed domestic AI drug discovery companies have formed three clearer ways of leading. Jitai Technology is moving fastest on registration progress. MTS-004 has completed Phase III clinical trials, and the licensor is advancing production validation and NDA filing preparation; MTS-201 has completed Phase I Part B; MTS-105 has entered dose escalation in a liver cancer IIT, and MTS-109 has obtained early human data and is preparing for U.S. and China INDs. Its core barrier lies in organ-targeted LNP, RNA sequence design, and in vivo expression, improving druggability efficiency by simultaneously designing the "payload" and the "delivery method." Moderna's mRESVIA went from the first elderly subject dosed in January 2021 to FDA approval in May 2024, a full cycle of about three years and four months. This timescale shows that after a mature mRNA platform completes process and regulatory validation, subsequent products have the opportunity to reuse the development foundation. If Jitai establishes an advantage through organ-targeted delivery, orphan drug designation, and early human data, MTS-105 may have the possibility of accelerated development, though the actual registration timeline still depends on clinical results. INSILICO has turned the small-molecule route into a scaled matrix: its official website discloses more than 40 projects and 13 INDs approved, and its core pipeline Rentosertib has launched Phase III for IPF. The study plans to enroll 320 people across 47 centers in China, with continuous dosing for 52 weeks, and a registration completion time of October 2029; enrollment, last-patient follow-up, and data cleaning explain the time gap of "one year of dosing, three years to complete." Phase III usually requires hundreds to thousands of subjects, with cycles measured in years. BIO statistics show that the success rate from Phase III to NDA/BLA across disease types is about 57.8%. IPF especially tests patient heterogeneity, background medication, and long-term lung function endpoints: ziritaxestat and pamrevlumab both released positive signals in Phase II, then were terminated in Phase III due to benefit-risk and primary endpoint issues. If Rentosertib can pass this major test, it will become an important validation for the entire AI drug discovery industry; the enrollment efficiency of Chinese clinical centers is expected to improve the time cost. XtalPi presents a third kind of leadership: the final form of its pipeline covers small molecules, antibodies, molecular glues, peptides, small nucleic acids, and cell therapy. Its 2026 interim report disclosed 3 clinical pipelines, more than 10 pipelines approved or preparing for IND, nearly 10 PCC pipelines, and more than 20 discovery-stage projects; by 2027, the goal is more than 10 clinical, more than 10 IND-stage, and about 20 PCC pipelines, meaning more than 40 pipelines reaching PCC or a more mature stage, forming the world's broadest drug modality pipeline. What better reflects the platform's advancement is that these pipelines are already distributed across key value nodes of different modalities: in small molecules, SIGX1094, RTX-117, and PEP08 have entered clinical trials, and SIGX2649 has received U.S. IND approval; in antibodies, 3 wholly owned pipelines are planned to enter clinical trials in 2027; of 6 siRNA pipelines, more than half have completed in vivo efficacy evaluation, and the fastest project has reached PCC; a molecular glue project obtained a picomolar-level degrader within one quarter; brain-delivery peptides and oral cyclic peptides are advancing toward PCC; and incubated cell therapy projects have obtained multiple U.S. and China IND clearances. Advancing six major drug modalities steadily to high maturity at the same time on the same technology foundation is almost unimaginable in the history of traditional biotech development. In the past, this was limited by the bottlenecks of a single technology platform; now, the market favors platforms with "system-level optionality," and more AI companies even prefer biologics modalities such as antibodies and small nucleic acidsthey allow the drug modality to actively adapt to biology and precisely match the most efficient weapon for a specific disease. It is precisely along this "optionality" logic that when we examine the final pipeline forms announced by each company, we find that the industry's division of labor and barriers have already shown clear differentiation. The table below counts only the final drug modalities that have been disclosed. Merely having algorithms or service capabilities is not counted as a formal pipeline; represents disclosed platform or collaboration capabilities, but no clearly named assets have yet been announced. Comparison of officially announced pipeline modalities of AI drug discovery companies As of public information in September 2026. The table counts self-developed, partnered, or incubated pipelines officially announced by the companies, and stage information is based on the companies' latest public statements. Table notes: clear assets or pipelines already exist; platform or collaboration capabilities disclosed, but no clearly named assets yet; no officially announced pipeline seen. Source: company announcements, official websites, and public clinical progress; compiled as of September 21, 2026. The industry division of labor revealed by the table is very clear: Jitai leads in registration progress and delivery technology, Insilico leads in the number of small-molecule pipelines and late-stage validation, AbCellera, Absci, and Generate focus on proteins and antibodies; XtalPi has the broadest disclosed modality portfolio. Multimodality is becoming the common choice for major players entering AI drug discovery Google DeepMind has extended AlphaFold to proteins, DNA, RNA, ligands, and antibody complexes; NVIDIA BioNeMo focuses on protein binder design; OpenAI is collaborating with Retro Biosciences to develop protein engineering models. After major players enter drug R&D, they generally start from large molecules and new modalities, because these fields have scarce data, more complex structural space, and better reflect the value of a computation-experimentation loop. Different targets correspond to different optimal solutions: intracellular pockets are suitable for small molecules, cell surface signaling can be blocked by antibodies, pathogenic proteins can be degraded through molecular glues, gene expression can be silenced by siRNA, and immune resetting can call on TCEs or CAR-T. What a multimodal platform gains is the optionality of "target first, modality later." Historical data also provides a reference. In BIO statistics on the probability from Phase I to approval, small molecules are about 7.5%, monoclonal antibodies about 12.1%, RNAi about 13.5%, and CAR-T about 17.3%. These cross-period data cannot directly predict a single pipeline, but they show that drug modality itself is a risk allocation tool. XtalPi's breadth comes from horizontal reuse of the same foundation The reason XtalPi's modalities can achieve the broadest layout is that they share a foundation based on quantum mechanics and physics-based modeling, AI generation and prediction, automated synthesis and experimental validation, and then standardized data flowing back into the model. The R&D workflow and the practical experience of "dry-wet integration" in AI drug discovery are fundamentally shared, and expansion speed is driven by platform reuse rate. The current small-molecule clinical portfolio includes SIGX1094, RTX-117, and PEP08; SIGX2649 has received U.S. IND approval and has submitted a China IND. The self-developed TRK/RET gut-restricted small molecule XTN004 has completed U.S. pre-IND material submission, with the interim report stating U.S. and China filings in the second half of 2026; according to recent company communications, the project is planned to submit an IND this week, subject to formal announcement. Large molecules and new modalities are also flourishing on multiple fronts. ALX001, ALX002, and ALX005 from XtalPi subsidiary Ailux are planned to enter clinical trials in 2027; Kodexia has laid out 6 siRNA pipelines, more than half of which have completed in vivo efficacy evaluation, among which the IgA nephropathy project obtained non-human primate data in about 7 months; a molecular glue project has obtained a picomolar-level degrader within one quarter; the peptide platform's brain-delivery and oral cyclic peptide projects are advancing toward PCC respectively; and incubated company Laimeng Biology's META 10-19 has obtained multiple U.S. and China IND clearances. SIGX1094 provides a sample for observing platform conversion efficiency. The project targets diffuse gastric cancer, observed preliminary safety and antitumor signals in Phase I, and received FDA orphan drug and fast track designations; its Phase II/III application for combination with Innovent's KRAS-G12C inhibitor has been accepted by the CDE. Fast track can improve regulatory communication and rolling review efficiency, and the subsequent registration path depends on Phase II design and efficacy strength. The pipeline is expected to obtain a rare disease Phase III exemption and could begin the marketing process as early as after completing Phase II clinical trials in 2027. Xige Shengke is responsible for clinical advancement, while XtalPi retains a share of subsequent commercialization revenue, allowing platform value to continue to be realized as the asset matures. In the first half of 2026, XtalPi's AI4S business revenue increased 136.4% year over year to RMB 193.5 million. The growth in pipeline numbers and multiple modalities simultaneously crossing PCC and IND nodes show that foundation reuse has begun to asset density. Ailux and DoveTree: the platform begins to actively pursue pipeline returns XtalPi subsidiary Ailux can be seen as a vanguard of its business model evolution. The company successively licensed its structure prediction platform to Johnson & Johnson and UCB, then reached a bispecific antibody collaboration with Eli Lilly worth up to US$345 million, with the contract including a platform licensing option, collaborative R&D, and milestone revenue; now it is further concentrating resources on three wholly owned autoimmune pipelines. This change quickly completed the value leap path of "platform licensingco-development and revenue sharingproprietary pipelines." The AtlaX proprietary data foundation contains billions of protein sequences, about 140,000 antigen-antibody sequence pairs, and about 30 million synthetic structure data points, and together with a 30,000-square-foot wet lab forms a solid feedback loop. While Anthropic is still filling in the experimental link by acquiring and building its own biological laboratory, Ailux has already accumulated real-world data that can directly serve model training. Maria Belvisi, who joined Ailux in April this year, was formerly senior vice president of respiratory and immunology R&D at AstraZeneca, managed about 500 scientists, and advanced tozorakimab from preclinical to Phase III. Her joining points to a clear goal: upgrading the AI antibody platform into a biotech with global clinical development capabilities. As a primary-market valuation reference, Anew Labs recently completed a US$290 million financing at a post-money valuation of about US$1.5 billion, while its fastest public pipeline is still preclinical. By contrast, Ailux already has multinational pharmaceutical clients, proprietary data, a wet lab, and three clear proprietary pipelines, with significantly higher asset maturity. The DoveTree collaboration shows that XtalPi began relatively early to bring in external leverage for clinical and commercialization capabilities. XtalPi has received a US$51 million upfront payment and a second payment of US$19 million, totaling US$70 million; the near-term payments remaining under the original agreement are up to US$30 million, with total potential value of up to US$5.99 billion, and the first oncology asset has entered the IND-enabling stage. DoveTree founder Gregory Verdine has co-founded more than ten biotechnology companies, of which more than five have listed on capital markets; Parabilis, renamed from Fog Pharma, listed in June this year. At signing, XtalPi disclosed that he had co-developed 3 FDA-approved drugs; the pancreatic cancer near-blockbuster daraxonrasib approved this year, whose RAS(ON) tri-complex technology engine absorbed early work by Verdine and Warp Drive, further increasing the approved-drug record of this technology lineage to 4. In this complementarity, XtalPi needs Verdine's seasoned judgment on targets and commercialization paths, while Verdine values XtalPi's hardcore ability to integrate AI, physical intelligence, and automated experimentation into the same platform. This combined force is turning the number of pipelines on paper into high-value assets in real money. The key to valuation: converting technological breadth into continuous value nodes In terms of speed, Jitai is closest to product registration, and Insilico has an AI-native small-molecule Phase III benchmark; in terms of the breadth of modalities and business models, XtalPi occupies a scarce position among listed AI drug discovery companies globally. This breadth can bring investors three layers of tangible defensive and offensive value: 1. Decentralized pipeline risk: more independent clinical and BD events reduce the degree to which the success or failure of a single pipeline dominates the company's overall valuation. 2. Biology-oriented "modality options": comparing different modalities around the same target allows the drug format to truly serve biological questions rather than being constrained by technological limitations. 3. Multi-level revenue structure: platform service revenue, milestones, sales royalties, proprietary pipelines, and incubated ecosystem equity build a rich profit pool. Traditional biotech valuations are often hostage to one or two core assets; XtalPi is closer to a composite valuation model of "R&D infrastructure revenue + risk-adjusted value of proprietary pipelines + partnered pipeline revenue sharing + options on Ailux and the incubated ecosystem + new materials platform." The current market has already priced its service revenue to some extent, but the valuation of its multimodal pipelines and vast ecosystem rights remains at a very early stage. The next value realization nodes are already clear: XTN004 formally submits its IND, SIGX1094 enters the next clinical stage, three ALX antibodies enter clinical trials in 2027, and small nucleic acid and peptide projects continue to reach PCC. The policy tailwind has raised the industry's risk appetite, but the scarcity of the underlying technology determines how far the premium can go. If secondary-market investors are betting that AI drug discovery is upgrading from a "single-point blind box tool" to an "industrialized system capable of repeatedly producing blockbuster drugs," then XtalPi, with the "broadest layout," undoubtedly has the most complete value mapping in the current market.