Q2 2026 XtalPi Holdings Ltd Earnings Call

August 19, 2026

Speaker #1: You have joined the meeting as an attendee and will be muted throughout the meeting.

Speaker #2: Hi everyone, and welcome to XtalPi Holdings' 2026 interim results presentation. All participants are currently on mute. We'll now read the following statement. This conference is for invited investors only.

Sun Choi Chung: Hi, everyone, and welcome to XtalPi Holdings 2026 Interim Results presentation. All participants are currently on mute. We will now read the following statement. This conference is for invited investors only. The audio and transcripts are for participants' internal use only and may not be published. XtalPi Holdings has not authorized any media to redistribute this conference. Unauthorized reproduction or redistribution is an infringement, and XtalPi Holdings reserve the right to take legal action. XtalPi Holdings accepts no liability for any loss arising from such reproduction or redistribution. Investing involves market risk. Please make investment decisions with care. Before we begin, please note that there will be time for questions after the speakers have finished their remarks. I will now invite management to begin their remarks. Thank you. Good evening and good morning, everyone. Welcome to XtalPi's 2026 Interim Results conference call. This is Sun Choi Chung, the company's IR.

Operator: Hi, everyone, and welcome to XtalPi Holdings 2026 Interim Results presentation. All participants are currently on mute. We will now read the following statement. This conference is for invited investors only. The audio and transcripts are for participants' internal use only and may not be published. XtalPi Holdings has not authorized any media to redistribute this conference. Unauthorized reproduction or redistribution is an infringement, and XtalPi Holdings reserve the right to take legal action. XtalPi Holdings accepts no liability for any loss arising from such reproduction or redistribution. Investing involves market risk. Please make investment decisions with care. Before we begin, please note that there will be time for questions after the speakers have finished their remarks. I will now invite management to begin their remarks. Thank you.

Speaker #2: The audio and transcript are for participants' internal use only and may not be published. XtalPi Holdings has not authorized any media to redistribute this conference.

Speaker #2: Unauthorized reproduction or redistribution is an infringement, and XtalPi Holdings reserves the right to take legal action. XtalPi Holdings accepts no liability for any loss arising from such reproduction or redistribution.

Speaker #2: Investing involves market risk. Please make investment decisions with care. Before we begin, please note that there will be time for questions after the speakers have finished their remarks.

Speaker #2: I'll now invite management to begin their remarks. Thank you. Good evening and good morning, everyone. Welcome to XtalPi's 2026 interim results conference call. This is Centralia Chung, the company's IR.

Sun Choi Chung: Good evening and good morning, everyone. Welcome to XtalPi's 2026 Interim Results conference call. This is Sun Choi Chung, the company's IR.

Speaker #2: On today's call, we have Dr. Xu Haowen, our co-founder, chairman of the board, and executive director, and Dr. Qian Ma, our co-founder, executive director, and chief executive officer.

Sun Choi Chung: On today's call, we have Dr. Shuhao Wen, our co-founder, Chairman of the Board, and Executive Director. Dr. Jian Ma, our co-founder, Executive Director, and Chief Executive Officer. Mr. Jeff Zhou, our CFO. Dr. Wen and Dr. Ma will provide an overview of our strategies and business development, and Mr. Zhou will provide additional detail on the company's financial results. With that, I will now turn the call over to our co-founder, Chairman of the Board, and our Executive Director, Dr. Wen. Please go ahead, Shu. Thank you. Welcome, everyone, to XtalPi's 2026 interim earnings presentation. Today, we will report our H1 2026 operating results and strategic progress. Before the numbers, I want to share our view of the AI for Science landscape and where XtalPi sits in this shift. That should make our business layout and growth logic easier to follow.

Speaker #2: And Mr. Jeff Zhou, our CFO. Dr. Wen and Dr. Ma will provide an overview of our strategies and business development, and Mr. Zhou will provide additional detail on the company's financial results.

Speaker #2: With that, I'll now turn the call over to our co-founder, Chairman of the Board, and Executive Director, Dr. Wen. Please go ahead, Zhou.

Speaker #2: Thank you. Welcome, everyone, to XtalPi's 2026 interim earnings presentation. Today, we'll report our first-half 2026 operating results and strategic progress. Before discussing the numbers, I want to share our view of the AI-for-science landscape, and where XtalPi sits in this shift.

Shuhao Wen: Thank you. Welcome, everyone, to XtalPi's 2026 interim earnings presentation. Today, we will report our H1 2026 operating results and strategic progress. Before the numbers, I want to share our view of the AI for Science landscape and where XtalPi sits in this shift. That should make our business layout and growth logic easier to follow.

Speaker #2: And that should make our business layout and growth logic easier to follow. We believe this round of technology shift presents great opportunities in front of us.

Shuhao Wen: We believe this round of technology shift presents great opportunities in front of us. The first wave of generative AI showed its productivity-enhancing value most clearly in AI coding. As you can see, it has created unprecedented high market cap companies in the market. In 2025 and 2026, leading figures in global AI have pointed to AI4S as the next stop. DeepMind co-founder Demis Hassabis says that we are at the door of a new golden age of scientific discovery. Former Google Chief Scientific Officer Jeff Dean has gone further, founding Discovery Loop to let AI close the experimental loop in science and engineering. His view is that AI's next major frontier will move from answering questions to creating discoveries. AI coding speeds up the production of existing knowledge. AI4S speeds up the creation of new knowledge, new drugs, new materials, and new processes.

Speaker #2: The first wave of generative AI showed its productivity-enhancing value most clearly in AI coding. As you can see, it has created unprecedented, high-market-cap companies in the market.

Speaker #2: In 2025 and 2026, leading figures in global AI have pointed to AI for us as the next stop. DeepMind co-founder Demis Hassabis says that we are at the door of a new golden age of scientific discovery.

Speaker #2: From Google Chief Scientific Officer Jeff Dean has gone further—founding Discovery Loop to let AI close the experimental loop in science and engineering.

Speaker #2: His view is that AI's next major frontier will move from answering questions to creating discoveries. AI coding speeds up the production of existing knowledge.

Speaker #2: AI for us speeds up the creation of new knowledge, new drugs, new materials, and new processes. Industry research suggests that the global AI for us market could exceed $140 billion USD.

Shuhao Wen: Industry research suggests that the global AI4S market could exceed USD 140 billion. We believe that AI4S can become the next high beta growth opportunity after AI coding. In fact, primary market interest is already very strong in this regard. Companies such as Lila Sciences, founded only recently, have attracted large-scale capital and valuations in the United States, approaching USD 10 billion since early founding days. Their core direction is very similar to XtalPi's, which is not just about building the AI models, it is combining generative AI with robotic labs to form a complete research loop. Objectively speaking, XtalPi has been doing this for over 10 years, and we have connected the digital and physical world, and we have already scaled this with leading global customers.

Shuhao Wen: Industry research suggests that the global AI4S market could exceed $140 billion. We believe that AI4S can become the next high beta growth opportunity after AI coding. In fact, primary market interest is already very strong in this regard. Companies such as Lila Sciences, founded only recently, have attracted large-scale capital and valuations in the United States, approaching $10 billion since early founding days. Their core direction is very similar to XtalPi's, which is not just about building the AI models, it is combining generative AI with robotic labs to form a complete research loop. Objectively speaking, XtalPi has been doing this for over 10 years, and we have connected the digital and physical world, and we have already scaled this with leading global customers.

Speaker #2: We believe that AI, for us, can become the next high-beta growth opportunity after AI coding. In fact, primary market interest is already very strong in this regard.

Speaker #2: Companies such as LiDA Science, founded only recently, have attracted large-scale capital and valuations in the United States, approaching $10 billion. Since the early founding days, their core direction is very similar to XtalPi's.

Speaker #2: This is not just about building AI models; it is about combining generative AI with robotic labs to form a complete research loop. Objectively speaking, XtalPi has been doing this for over 10 years. We have connected the digital and physical worlds, and we have already scaled this with leading global customers.

Speaker #2: Therefore, as one of the few listed companies in this field, we have a clear first-mover advantage and an industry-validated foundation. We are well-placed to benefit from this industry-level opportunity.

Shuhao Wen: Therefore, as one of the few listed companies in this field, we have a clear first-mover advantage and also an industry-validated foundation. We are well-placed to benefit from this industry-level opportunity. We believe that we also have great opportunity to become the next beta growth opportunity. Based on this practice, we think that the core of AI for science is not simply stacking models, robots, and computation. Rather, it is building intelligent research infrastructure that runs through the digital and physical worlds. AI can move from an assistive tool to autonomous scientific discovery. That means AI no longer only completes local tasks on human instruction. Instead, it can organize work around a scientific goal, make specialist decisions, call experiments for validation, and keep iterating based on the results. Autonomous scientific discovery advances level by level. We have taken the lead in defining five levels.

Speaker #2: And we believe that we also have a great opportunity to become the next beta growth opportunity. Based on this practice, we think that the core of AI for science is not simply stacking models, robots, and computation.

Speaker #2: Rather, it is building intelligent research infrastructure that runs through the digital and physical worlds, so AI can move from an assistive tool to autonomous scientific discovery.

Speaker #2: That means AI no longer only completes local tasks on human instruction. Instead, it can organize work around a scientific goal, make specialist decisions, call experiments for validation, and keep iterating based on the results.

Speaker #2: Autonomous scientific discovery advances level by level. We have taken the lead in defining five levels: L1 is tool use, L2 is assisting scientists, L3 is agents independently completing parts of a workflow, and L4 is full workflow autonomy.

Shuhao Wen: L1 is tool use, L2 is assisting scientists, L3 is agents independently completing parts of a workflow, L4 is full workflow autonomy in a specific domain, and L5 is general autonomous scientific discovery across domains. We have already achieved end-to-end L4 capability in several concrete scenarios. The capability sits on a four-layer technology architecture that we have refined over many years. First is Genius Agents, the intelligent hub. Think of it as the brain and dispatch system of scientific discovery. It has a large set of specialist skills and tools. It can break down complex scientific goals, orchestrate models, tools, R&D workflows, and data, and coordinate multiple specialist agents on long-running tasks. The second level is, or the second layer is scientific AI. Specialist models built around concrete scientific problems.

Speaker #2: In a specific domain, L5 is general autonomous scientific discovery across domains. And we have already achieved end-to-end L4 capability in several concrete scenarios.

Speaker #2: Now, the capability sits on a four-layer technology architecture that we have refined over many years. First is Genius Agent, the intelligent hub. Think of it as the brain and dispatch system of scientific discovery.

Speaker #2: It has a large set of specialist skills and tools. It can break down complex scientific goals, orchestrate models, tools, R&D workflows, and data, and coordinate multiple specialist agents on long-running tasks.

Speaker #2: The second level, or the second layer, is scientific AI. Specialist models are built around concrete scientific problems. Across small molecules, large molecules, peptides, and oligonucleotides, we have launched a series of proprietary models and platforms covering molecular design, structure prediction, function assessment, and candidate optimization.

Shuhao Wen: Across small molecules, large molecules, peptides, and oligonucleotides, we have launched a series of proprietary models and platforms covering molecular design, structure prediction, function assessment, and candidate optimization. Third is physical AI. Our self-built large-scale robotic laboratories turn AI-generated experimental plans into standardized, automated, and traceable workflows that robots can execute. Fourth is our data moat, a high-quality data foundation. Scientific AI ultimately depends on data quality. We combine public data, proprietary data, and real experimental data from physical AI into traceable data assets. Our physical AI system now stably produces more than 50,000 reaction yield data points and 300,000 processed data points each month. We have accumulated more than 500,000 complete real experimental records. Around 80% are failed negative samples, which are very scarce in published literature. These negative samples are critical for training reliable models, identifying unproductive paths, and calibrating prediction boundaries.

Speaker #2: Third is physical AI. Our self-built, large-scale robotic laboratories turn AI-generated experimental plans into standardized, automated, and traceable workflows that robots can execute. Fourth is our data modes—a high-quality data foundation.

Speaker #2: Scientific AI ultimately depends on data quality. We combine public data, proprietary data, and real experimental data from Physical AI into traceable data assets. Our Physical AI system now stably produces more than 50,000 reaction yield data points and 300,000 process data points each month.

Speaker #2: We have accumulated more than 500,000 complete real experimental records. Around 80% are failed negative samples, which are very scarce in published literature. These negative samples are critical for training reliable models, identifying unproductive paths, and calibrating prediction boundaries.

Speaker #2: On this foundation, we have a dual-engine business: Discovery—the drug discovery solutions on one side, and AI for Science, or AI for US intelligent solutions, on the other.

Shuhao Wen: On this foundation, we have a dual-engine business. The drug discovery solutions on one side, and AI for Science or AIfS intelligent solutions on the other. Drug discovery solutions include platform collaboration services and licensing of proprietary assets. Service contribute relatively stable cash flow as FBD offers higher upside revenue. AI for Science intelligence solutions provide customers with AI for Science infrastructure, including robotic labs and intelligence services. These two businesses are not separate from each other. They reinforce each other. Drug discovery keeps generating data and validation needs, which upgrades our AI for Science capabilities. The shared technology foundation from AI for Science then gives drug discovery a more efficient R&D infrastructure that create a recursive improvement loop, scenario demand, experimental validation, data feedback, and capability improvement. Our platform is also extending into advanced materials and consumer health, showing that it can be reused across industries.

Speaker #2: Drug discovery solutions include platform collaboration services and licensing of proprietary assets. Services contribute relatively stable cash flow. Asset BD offers higher upside revenue. AI-for-Science intelligent solutions provide customers with AI-for-Science infrastructure, including robotic labs and intelligent services.

Speaker #2: These two businesses are not separate from each other; they reinforce each other. Drug discovery keeps generating data and validation needs, which upgrades our AI for science capabilities.

Speaker #2: The shared technology foundation from AI for science then gives drug discovery a more efficient R&D infrastructure that creates a recursive improvement loop scenario. Demand, experimental validation, data feedback, and capability improvement platform is also extending into advanced materials and consumer health, showing that it can be reused across industries.

Speaker #2: And we delivered several business highlights in the first half. You can see that many of our customers have actually generated good repurchase rates.

Shuhao Wen: We delivered several business highlights in H1. You can see a lot of our customers actually has generated good repurchase rates. For H1 2026, firstly, in AI for Science infrastructure, we are the only player with a full stack system connecting robotic labs, scientific agents, and autonomous synthesis, and we have already commercialized it at scale. In the reporting period, AI for Science intelligent solutions posted a very strong growth with revenue up 136.4% year-over-year. This is very strong. Second, in our drug R&D, we have built four core platforms covering small molecules, large molecules, peptides, and oligonucleotides, one of the broadest modality footprints in the industry. Each platform has high-quality, standardized proprietary data and industry-leading generative and predictive models. This is our core technology foundation. Third, on the pipeline, we, compared to last year, have made great progress.

Speaker #2: Now, for the first half of 2026—firstly, in AI-for-science infrastructure, we are the only player with a full-stack system connecting robotic labs, scientific agents, and autonomous synthesis.

Speaker #2: And we have already commercialized it at scale. In the reporting period, AI for Science intelligent solutions posted very strong growth, with revenue up 136.4% year over year.

Speaker #2: This is very strong. Second, in our drug R&D, we have built four core platforms covering small molecules, large molecules, peptides, and oligonucleotides—one of the broadest modality footprints in the industry.

Speaker #2: Each platform has high-quality, standardized proprietary data and industry-leading generative and predictive models. This is our core technology foundation. Third, on the pipeline, compared to last year, we have made great progress.

Speaker #2: We have efficiently built a rich and highly differentiated mix of partner and proprietary programs. So far, three programs are in clinical trials. More than 10 are IND approved or in IND-enabling stages.

Shuhao Wen: We have efficiently built a rich and highly differentiated mix of partnered and proprietary programs. So far, three programs are in clinical trials. More than 10 are IND-approved or in IND-enabling. Nearly 10 have reached PCC. Looking into 2027, we expect more than 10 programs in clinical trials, more than 10 at IND-approved or IND-enabling, and around 20 at PCC. That is very fast-paced for the industry. Here, we want to stress that our company is prioritizing the quality of the pipeline and not just solely focused on speed. Recently, you already have heard that many pipelines that we successfully incubated and helped developed, have received industry recognition. The DIANA Award, which was also named the Nobel Prize of the industry. Fourth, I want to stress that our AI is production-grade. It has already broken traditional R&D bottlenecks across modalities.

Speaker #2: Nearly 10 have reached PCC. Looking into 2027, we expect more than 10 programs in clinical trials, more than 10 at IND-approved or IND-enabling, and around 20 at PCC.

Speaker #2: That is a very fast pace for the industry. But here, we want to stress that our company is prioritizing the quality of the pipeline and not just solely focusing on speed.

Speaker #2: So recently, you already have heard that many pipelines that we successfully incubated and helped develop have received industry recognition—the DIA award, which was also named the Nobel Prize of the Industry.

Speaker #2: Fourth, I want to stress that our AI is production-grade. It has already broken traditional R&D bottlenecks across modalities. In antibody programs, we achieved mechanism innovation and treatment window optimization.

Shuhao Wen: In antibody programs, we achieved mechanism innovation and treatment window optimization. Some molecular glue programs reached pico mole target protein degradation within one quarter. This is definitely a phenomenal progress. An oral cyclic peptide program found its compound just two months after target selection. An oligonucleotide program for IgA nephropathy delivered excellent non-human primate efficacy data about seven months after launch. These examples show that our AI can truly move programs forward and drive innovative R&D. A lot of these high-quality pipelines are based on our core innovative research and development capability. Fifth, we continue to deepen our partnerships with leading global customers. The work has expanded from single services to platform services, model licensing, pipeline transactions, and AI for Science infrastructure deployment. Overseas delivery and repeat orders from top customers validate the strength of our business model.

Speaker #2: Some molecular glue programs reached picomolar target protein degradation within one quarter, and this is definitely phenomenal progress. Also, the cyclic peptide program found its compound just two months after target selection, and an oligonucleotide program for IgA nephropathy delivered excellent non-human primate efficacy data about seven months after launch.

Speaker #2: These examples show that our AI can truly move programs forward and drive innovative R&D. Many of these high-quality pipelines are based on our core innovative research and development capabilities.

Speaker #2: Fifth, we continue to deepen our partnerships with leading global customers. The work has expanded from single services to platform services, model licensing, pipeline transactions, and AI-for-science infrastructure deployment.

Speaker #2: Overseas delivery and repeat orders from top customers validate the strength of our business model. Sixth, on technology, we launched Shoreld, our self-improving AI retrosynthesis system.

Shuhao Wen: Sixth, on technology, we launched SureRoute, our self-improving AI retrosynthesis system. It cut the chemical hallucination rate to 4.6%, only one-sixth of leading large models in the industry. Our large model has the lowest hallucination rates in the industry. Top ranked route accuracy reached 74.3%, 2 to 3.5 times that of existing specialists in general purpose models. This significantly improves the reliability of AI-assisted route design and the efficiency of R&D translation. Seven, we officially launched the XtalPi Science platform and the Genius Agents Suite. This is the world's first comprehensive open research platform to close the Physical AI loop. It gives global industry partners and research institutions on-demand access to underlying AI for Science infrastructure. Finally, in biological simulation, we have used investment and incubation to build capabilities in virtual cells, human organ-on-a-chip systems, and organoids.

Shuhao Wen: Sixth, on technology, we launched SureRoute, our self-improving AI retrosynthesis system. It cut the chemical hallucination rate to 4.6%, only one-sixth of leading large models in the industry. Our large model has the lowest hallucination rates in the industry. Top ranked route accuracy reached 74.3%, 2x to 3.5x that of existing specialists in general purpose models. This significantly improves the reliability of AI-assisted route design and the efficiency of R&D translation. Seven, we officially launched the XtalPi Science platform and the Genius Agents Suite. This is the world's first comprehensive open research platform to close the Physical AI loop. It gives global industry partners and research institutions on-demand access to underlying AI for Science infrastructure. Finally, in biological simulation, we have used investment and incubation to build capabilities in virtual cells, human organ-on-a-chip systems, and organoids.

Speaker #2: It cut the chemical hallucination rate to 4.6%, only one-sixth of leading large models in industry. So, our large model basically has the lowest hallucination rates in the industry.

Speaker #2: And top-ranked accuracy reached 74.3%, two to two and a half times that of existing specialists in general-purpose models. This significantly improves the reliability of AI-assisted route design and the efficiency of R&D translation.

Speaker #2: Seventh, we officially launched the Expo XtalPi Science Platform and the Genius Agent Suite. This is the world's first comprehensive open research platform to close the physical AI loop.

Speaker #2: It gives global industry partners and research institutions on-demand access to underlying AI-for-science infrastructure. Finally, in biological simulation, we have used investment and incubation to build capabilities in virtual cells, human organ-on-a-chip systems, and organoids.

Speaker #2: This extends our R&D from molecular design into mechanism research, translational prediction, and efficacy validation at the cell and tissue level, which should further improve preclinical translation success.

Shuhao Wen: This extends our R&D from molecular design into mechanism research, translational prediction, and efficacy validation at the cell and tissue level, which should further improve preclinical translation success. Overall, H1 progress both validated the leadership of our technology platform and strengthened the commercial foundation for future growth. We see stronger visibility and certainty of XtalPi's yearly growth. Looking ahead, AI for Science and AIDD in particular, is in a high-growth phase. With a full chain layout from R&D infrastructure to proprietary assets, we are well-placed to benefit. In the near term, rapid growth in AI drug discovery will drive more demand for new molecule synthesis and high-quality experimental data. With our AI-native experimental system and leading intelligent robotic labs, we can already take incremental orders from major pharma companies and research institutions supporting faster AI for Science growth.

Speaker #2: Overall, first-half progress above validated the leadership of our technology platform and strengthened the commercial foundation for future growth. And we see stronger visibility and certainty of XtalPi's yearly growth.

Speaker #2: Looking ahead, AI for science—and AIGD in particular—is in a high-growth phase, with a full-chain layout from R&D infrastructure to proprietary assets.

Speaker #2: We are well placed to benefit. In the near term, rapid growth in AI drug discovery will drive more demand for new molecule synthesis and high-quality experimental data.

Speaker #2: With our AI-native experimental system and leading intelligent robotics labs, we can already take incremental orders from major pharma companies and research institutions, supporting faster AI for science growth.

Speaker #2: Over the medium term, demand for innovative pipelines remains strong among pharma companies in China and overseas. Our drug discovery platform can continue generating revenue through platform services and can also unlock higher upside value through SFBD.

Shuhao Wen: Over the medium term, demand for innovative pipelines remains strong among pharma companies in China and overseas. Our drug discovery platform can keep generating revenue through platform services and can also unlock higher upside value through FMBD. This gives us two growth engines: platform services and asset monetization. Over the medium to long term, as proprietary pipelines move into filing and clinical trials, we will gradually form a dual engine model of R&D services plus proprietary innovative drugs, opening growth beyond traditional services. Over the long term, AI for Science is shifting pharmaceutical R&D from empirical trials and error to data and intelligent-driven development. By continuing to accumulate exclusive data, algorithms, and experimental capability in real projects, we will keep strengthening our platform barriers and be well-placed to capture long-term opportunities from this industry shift.

Speaker #2: This gives us two growth engines, platform services and asset monetization. Over the medium to long term, as proprietary pipelines move into filing and clinical trials, we'll gradually move well, we'll gradually form a dual engine model of R&D services plus proprietary innovative drugs, opening growth beyond trans you know, traditional services.

Speaker #2: Over the long term, AI for science is shifting pharmaceutical R&D from empirical trial and error to data- and intelligence-driven development. By continuing to accumulate exclusive data, algorithms, and experimental capability in real projects, we'll keep strengthening our platform barriers and be well placed to capture long-term opportunities from this industry shift.

Speaker #2: And so, AI is a great opportunity for the whole industry, and XtalPi is in a good position to benefit. I'll stop here. I'm now going to hand over to Dr. Atumajian to discuss our performance and business progress.

Shuhao Wen: AI is a great opportunity for the whole industry, and XtalPi is in a good position to benefit. I will stop here. I am going to now hand over to Dr. Ma Jian to discuss our performance and business progress. Thank you, Shuhao. Hi, everyone. I am Ma Jian. Coming up, allow me to discuss the business performance and business progress of XtalPi. In H1 2026, the group generated a revenue of 394 million RMB. Excluding the impact of a large pipeline BD, revenue grew 73.8% year-over-year. AI for Science intelligent solutions grew 136.4% year-over-year, with both robotic labs and intelligence services maintaining high growth. On profitability, H1 net loss was 225 million RMB, with an adjusted net loss of 106 million RMB. The wider loss mainly reflected deliberate R&D investment.

Speaker #2: Thank you, Shuhao. Hi, everyone. I'm Majin. Coming up, allow me to discuss the business performance and progress of XtalPi in the first half of 2026.

Jian Ma: Thank you, Shuhao. Hi, everyone. I am Ma Jian. Coming up, allow me to discuss the business performance and business progress of XtalPi. In H1 2026, the group generated a revenue of 394 million RMB. Excluding the impact of a large pipeline BD, revenue grew 73.8% year-over-year. AI for Science intelligent solutions grew 136.4% year-over-year, with both robotic labs and intelligence services maintaining high growth. On profitability, H1 net loss was 225 million RMB, with an adjusted net loss of 106 million RMB. The wider loss mainly reflected deliberate R&D investment.

Speaker #2: The group generated revenue of RMB 394 million. Excluding the impact of a large pipeline BD, revenue grew 73.8% year over year. AI for Science and Intelligent Solutions grew 136.4% year over year, with both Robotic Labs and Intelligent Services maintaining high growth.

Speaker #2: On profitability, first half net loss was RMB 225 million, with an adjusted net loss of RMB 106 million. The wider loss mainly reflected deliberate R&D investment.

Speaker #2: R&D expenses rose 66% year over year, focused on autonomous labs, agentic systems, pipeline programs, and multimodal platforms. These are strategic investments in long-term competitive barriers construction and asset value.

Jian Ma: R&D expenses rose 66% year-over-year, focused on autonomous labs, Agentic systems, pipeline programs, and multi-model platforms. These are strategic investments in long-term competitive barriers construction and asset value. On liquidity, as of 30 June, the group had a total cash balance of about 8.67 billion RMB. This reserve gives us ample support for continued intensive R&D and business expansion. In fact, excluding the impact of large pipeline deal, the amount of adjusted loss in the H1 was basically flat. The adjusted loss ratio narrowed sharply by 58%. Operating quality continues to improve. Now, coming up, I am going to elaborate on our business progress. First, let us talk about the drug discovery solutions. It generated 200 million RMB in the H1. The year-over-year decline mainly reflected the large pipeline licensing impact.

Speaker #2: On liquidity, as of June 30th, the group had a total cash balance of about RMB 8.67 billion. This reserve gives us ample support to continue intensive R&D and business expansion.

Speaker #2: In fact, excluding the impact of the large pipeline deal, the amount of adjusted loss in the first half was basically flat. But the adjusted loss ratio now drops sharply by 58%.

Speaker #2: Operating quality continues to improve. Now, coming up, I'm going to elaborate on our business progress. First, let's talk about the drug discovery solutions. It generated RMB 200 million in the first half.

Speaker #2: The year-over-year decline mainly reflected the large pipeline licensing impact. Importantly, we are advancing a strategic shift from services to assets. By building proprietary, innovative pipelines and multimodal platforms, we continue to accumulate assets with high clinical value.

Jian Ma: Importantly, we are advancing a strategic shift from services to assets by building proprietary innovative pipelines and multi-model platforms. We continue to accumulate assets with high clinical value. Short-term revenue fluctuation is a temporary base effect. It has not changed the company's long-term value. Now in drug R&D, we have built four core platforms covering small molecules, large molecules, peptides, and oligonucleotides, one of the most complete AI drug discovery modality footprints in industry. The core advantages start with industry-leading models, robotic experimental platforms, a distinctive data foundation, and an integrated biological simulation platform. More importantly, these capabilities are not just tools. They have long been deeply combined with real pharmaceutical scenarios. Our algorithm and wet lab teams work closely together around concrete drug discovery problems. We deliberately accumulate proprietary data sets and build robotic workstations designed for high-throughput data generation.

Speaker #2: Short-term revenue fluctuation is a temporary, you know, base effect. It does not change the company's long-term value. Now, in drug R&D, we have built four core platforms covering small molecules, large molecules, peptides, and oligonucleotides.

Speaker #2: One of the most complete AI drug discovery modality footprints in the industry. The core advantages start with industry-leading models, robotic experimental platforms, a distinctive data foundation, and an integrated biological simulation platform.

Speaker #2: More importantly, these capabilities are not just tools. They have long been deeply combined with real pharmaceutical scenarios. Our algorithm and wet lab teams work closely together around concrete drug discovery problems.

Speaker #2: We deliberately accumulate proprietary data sets and build robotic workstations designed for high-throughput data generation. This has given us a stronger data moat and know-how.

Jian Ma: This has given us a stronger data moat and know-how in selected R&D steps and modalities. As these capabilities keep accumulating across modalities, we continue to improve candidate discovery efficiency, development ability, optimization, and pipeline translation certainty. This lays a solid foundation for a more diversified innovative drug portfolio and external collaborations. In small molecules, I want to highlight our molecular glue platform. We have built systematic platform capability. We actually have accumulated more than 40 novel CRBN binding cores and 150,000 molecular building blocks in a virtual molecular glue library of about 2.58 million molecules. We have also combined AI design with robotic labs, automatically synthesizing and testing around 600 to 1,000 compounds a week. We have laid out multiple molecular glue programs in autoimmune disease and obtained hit compounds against multiple targets.

Speaker #2: In selected R&D steps and modalities, as these capabilities keep accumulating across modalities, we continue to improve candidate discovery efficiency, developability optimization, and pipeline translation certainty.

Speaker #2: This lays a solid foundation for a more diversified, innovative drug portfolio and external collaborations. In small molecules, I want to highlight our molecular glue platform.

Speaker #2: We have built systematic platform capability. We have actually accumulated more than 40 novel CRBM binding cores, 150,000 molecular building blocks, and a virtual molecule glue library of about 2.58 million molecules.

Speaker #2: We've also combined AI design with robotic labs, automatically synthesizing and testing around 600 to 1,000 compounds a week. We have laid out multiple molecular glue programs in autoimmune disease and obtained hit compounds against multiple targets.

Speaker #2: Some programs were able to optimize target protein degradation to picomolar concentrations in about one quarter, showing the platform's real project efficiency and translation potential.

Jian Ma: Some programs were able to optimize target protein degradation to picomolar concentrations in about one quarter, showing the platform's real project efficiency and translation potential. Now, on product pipeline, small molecule programs with several collaborators are progressing well with multiple products in clinical trials or at IND-related stages. These include Signet Therapeutics' dual FAK/SRC inhibitor, Revir Therapeutics' eIF2B activator, and PharmaEngine's PRMT5 inhibitor, all already in clinical trials. Signet Therapeutics and TEAD inhibitor has received IND clearance in China and the US. META Pharmaceuticals' LDH inhibitor and the DoveTree pan-cancer assets are also moving forward. Regarding large molecules, our core platform is Ailux, an AI-native antibody platform built on three AI engines. XtalFold understands and predicts protein structure. StemFOLD generates new molecular designs. Xentient judges how these designs perform on functions, activity, and developability. Beyond these models, a key differentiator is our proprietary foundational data.

Speaker #2: Now, on the partner pipeline, small molecule programs with several collaborators are progressing well, with multiple products in clinical trial or at IND-related stages. These include Signet Therapeutics' dual FAK/SRC inhibitor, Revyar Therapeutics' EF2B activator, and SEngine's PRMT5 inhibitor, all already in clinical trials.

Speaker #2: Signet Therapeutics' TEAD inhibitor has received IND clearance in both China and the US. Metapharmaceuticals' LDH inhibitor and the Dove Tree pan-cancer assets are also moving forward.

Speaker #2: Regarding large molecules, our core platform is AI Lux and our AI-native antibody platform, built on three AI engines. Xtofold understands and predicts protein structure.

Speaker #2: Semprod generates new molecular designs. Sentient judges how these designs perform on functional activity and developability beyond these models. Our key differentiator is our proprietary foundational data.

Speaker #2: Antibody development depends heavily on high-quality data. On key data types such as antibody affinity and antigen-antibody complex structures, our proprietary database has a several-fold to tenfold scale advantage over public datasets.

Jian Ma: Antibody development depends heavily on high-quality data on key data types such as antibody affinity and antigen antibody complex structures. Our proprietary database has a several-fold to tenfold scale advantage over public datasets. That means our models are not trained on only public data, that they keep absorbing proprietary data from real R&D, which creates a lasting barrier to entry. Combining these capabilities, Ailux has a relatively complete large molecule model system validated by more than 100 internal and external projects. Regarding our proprietary pipelines, we are steadily advancing three core large molecule programs focused on autoimmune disease. All three are expected to enter phase I in 2027. They are AI-driven, not only to help find candidates faster, but to systematically optimize the targeted mechanism, treatment window, safety, and developability. We believe each potentially has a best-in-class categorization.

Speaker #2: That means our models are not trained on only public data. As you know, they keep absorbing proprietary data from real R&D, which creates a lasting barrier to entry.

Speaker #2: Combining these capabilities, Alex has a relatively complete large molecule model system, validated by more than 100 internal and external projects. Regarding our proprietary pipelines, we are steadily advancing three core large molecule programs.

Speaker #2: Focused on autoimmune disease. All three are expected to enter Phase 1 in 2027. They are AI-driven, not only to help find candidates faster but to systematically optimize the targeted mechanism, treatment window, safety, and developability.

Speaker #2: We believe each potential has a best-in-class categorization. In peptides, our platform mainly addresses traditional bottlenecks: limited stability, limited cell membrane penetration, and low oral bioavailability.

Jian Ma: In peptides, our platform mainly addresses traditional bottlenecks: limited stability, limited cell membrane penetration, and low oral bioavailability. We have built a database of more than 5,000 non-standard amino acids and developed AI models for designing complex peptides such as cyclic peptides. We are already seeing progress. A brain discovery program has entered in vivo animal testing. The goal is to use cyclic peptides to help drugs cross the blood-brain barrier and deliver therapeutic molecules that would otherwise struggle to reach the brain. We expect this program to actually reach PCC in H1 2027. Another of our oral cyclic peptide program moved very fast, finding hit compounds about two months after launch. Optimization of the peptide is underway, and we expect to confirm PCC by mid-2027. Beyond drug pipeline, we also see faster landing opportunities in consumer health.

Speaker #2: We have built a database of more than 5,000 non-standard amino acids and developed AI models for designing complex peptides, such as cyclic peptides.

Speaker #2: We are already seeing progress. A brain discovery program has entered in vivo animal testing. The goal is to use cyclic peptides to help drugs cross the blood-brain barrier and deliver therapeutic molecules.

Speaker #2: That would otherwise struggle to reach the brain. We expect this program to actually reach PCC in the first half of 2027. Another of our oral cyclic peptide programs moved very fast, finding hit compounds about two months after launch.

Speaker #2: Optimization of the peptide is underway, and we expect to confirm PCC by mid-2027. Beyond our drug pipeline, we also see faster landing opportunities in consumer health for blood glucose management.

Jian Ma: For blood glucose health management, we developed an innovative oral food ingredient that has received a US GRAS status, meaning it is considered safe for use as a food-grade ingredient. In mouse models, the molecule achieved about a 30% reduction in blood glucose within two weeks, with 5% to 10% weight loss. We are now in discussions with several institutions to advance application in blood glucose health management. In oligonucleotides, our platform's focus is using AI to design nucleic acid sequences and chemical modifications together. Traditional methods often rely on fixed templates and screening within a limited range. Our SI-former architecture jointly generates and optimizes sequence and modification. That creates a chance to generate molecules with best-in-class potential, and it is also more favorable for a differentiated patent estate. More importantly, this architecture improves generalization to new targets.

Speaker #2: We developed an innovative oral food ingredient that has received US GRAS status, meaning it is considered safe for use as a food-grade ingredient.

Speaker #2: In mouse models, the molecule achieved lower blood glucose within two weeks, with 5% to 10% weight loss. We're now in discussions with several institutions for its advanced application.

Speaker #2: In blood glucose health management, in oligonucleotides, our platform's focus is using AI to design nucleic acid sequences and chemical modifications together. Traditional methods often rely on fixed templates and screening within a limited range.

Speaker #2: Our SI Former architecture jointly generates and optimizes sequence and modification. That creates a chance to generate molecules with best-in-class potential, and it is also more favorable for a differentiated patent estate.

Speaker #2: More importantly, this architecture improves generalization to new targets. Even for targets with less historical data and less experience, the platform can generate effective candidates more flexibly and efficiently.

Jian Ma: Even for targets with less historical data and less experience, the platform can generate effective candidates more flexibly and efficiently. Compared to traditional paths, our oligonucleotides platform has more than doubled R&D efficiency, and molecular property prediction accuracy has improved by about 266%. That is critical for shortening early discovery and improving decision quality. The oligonucleotide platform now has six programs covering IgA nephropathy, metabolism, and CNS. The most advanced has reached the PCC stage. The IgA nephropathy program obtained excellent non-human primate efficacy data about seven months after launch, with activity and durability better than a same target clinical stage reference molecule. Now, the platform capabilities I just described ultimately create real program progress and asset conversion. In this stage, we are showing XtalPi's pipeline layout to you for the first time. This table lists some of our pipeline programs at the PCC stage and beyond.

Speaker #2: Compared to traditional paths, our oligonucleotides platform has more than doubled R&D efficiency. And molecular property prediction accuracy has improved by about 266%. That is critical for shortening early discovery and improving decision quality.

Speaker #2: The oligonucleotide platform now has six programs covering IgA, nephropathy, metabolism, and CNS. The most advanced has reached the PCC stage. The IgA nephropathy program obtained excellent non-human primate efficacy data about seven months after launch, with activity and durability better than a same-target clinical-stage reference molecule.

Speaker #2: Now, the platform capabilities I just described ultimately create real program progress and asset conversion. So, at this stage, we are showing XtalPi's pipeline layout to you for the first time.

Speaker #2: This table lists some of our pipeline programs at the PCC stage and beyond. We have many earlier-stage programs that are not on this slide.

Jian Ma: We have many earlier stage programs that are not on this slide. As they advance, we will introduce them to investors later. On overall numbers, across proprietary and enabled pipelines, three programs are now in the clinical stage. More than 10 are in IND approved or in IND preparation, and nearly 10 have reached the PCC stage. Looking to 2027, we expect more than 10 in the clinical stage, more than 10 at IND approval or IND preparation, and around 20 at the PCC stage. The portfolio covers oncology, autoimmune disease, metabolism, chronic disease, neurology, and consumer health. The potential market is in the hundreds of billions of USD, with very broad commercialization potential. The speed of pipeline progress shows that we have built a highly efficient R&D platform. In 2025, we had fewer than 10 programs.

Speaker #2: As they advance, we'll introduce them to investors later. On overall numbers, across proprietary and enabled pipelines, three programs are now in the clinical stage.

Speaker #2: More than 10 are in IND-approved or in IND preparation, and nearly 10 have reached the PCC stage. Looking to 2027, we expect more than 10 in the clinical stage.

Speaker #2: More than 10 at IND approval or IND preparation, and around 20 at the PCC stage. The portfolio covers oncology, autoimmune disease, metabolism, chronic disease, neurology, and consumer health.

Speaker #2: The potential market is in the hundreds of billions of US dollars, with very broad commercialization potential. The speed of pipeline progress shows that we have built a highly efficient R&D platform.

Speaker #2: In 2025, we have fewer than 10 programs. In just two years, by 2027, we expect to advance nearly 40 programs to the PCC stage and beyond.

Jian Ma: In just two years, by 2027, we expect to advance nearly 40 programs to the PCC stage and beyond. Platform capability is rapidly converting into visible pipeline assets. The key reason is that we have always embedded AI deeply in real pharmaceutical scenarios, rather than stopping at models or concepts. Through continuous iteration of data, models, and experimental closed loops, we can discover and optimize candidates faster, and we also have a chance to raise program certainty and success probability. On commercialization, we had a dense set of BD progress in H1 2026, further validating external recognition of the platform and its monetization potential. First, we enter an AI drug discovery strategic collaboration with a well-known international pharmaceutical company with a total potential value of more than $400 million. The two sides will develop all those small molecules against GPCR targets.

Speaker #2: Platform capability is rapidly converting into visible pipeline assets. The key reason is that we have always embedded AI deeply in real pharmaceutical scenarios, rather than stopping at models or concepts.

Speaker #2: Through continuous iteration of data, models, and experimental closed loops, we can discover and optimize candidates faster. And we also have a chance to raise programs—you know, raise programs’ certainty and success probability.

Speaker #2: On commercialization, we had a dense set of BD progress in the first half of 2026, further validating external recognition of the platform and its monetization potential.

Speaker #2: First, we entered an AI drug discovery strategic collaboration with a well-known international pharmaceutical company, with a total potential value of more than $400 million.

Speaker #2: The two sides will develop oral small molecules against GPCR targets. This is a typical innovative drug BD structure. The partner pays an upfront fee and funds early R&D.

Jian Ma: This is a typical innovative drug BD structure. The partner pays an upfront fee and funds early R&D. If programs progress, we will also receive milestones and sales royalties. Second, we continue to explore deeper platform co-built models. We plan to form a joint venture with Sunshine Lake Pharma. The partner is expected to invest several hundreds of millions of RMB to jointly build an AI plus robotics lab and a foundational model for preclinical trials, preclinical drugs, creating a model of pipeline co-creation plus share technology. We have also entered a collaboration worth tens of millions of RMB with Bison Pharma focused on endocrine and metabolic disease. The DoveTree collaboration is also progressing well. We have already received a second payment of $19 million USD. Our AI peptide projects with Gan & Lee Pharmaceuticals was also recognized as a Beijing Key Lab.

Speaker #2: If programs progress, we will also receive milestone and sales royalties. Second, we continue to explore a deeper platform called Built Models. We plan to form a joint venture with Sunshine Lake Pharma.

Speaker #2: The partner is expected to invest several hundreds of millions of R&D—I mean, R&B—to jointly build an AI plus robotics lab and a foundational model for preclinical trials, preclinical drugs, creating a model of pipeline co-creation plus shared technology.

Speaker #2: We've also entered a collaboration worth tens of millions of RMB with Vison Pharmaceuticals, focused on endocrin and metabolic disease. The DOVE3 collaboration is also progressing well, and we have already received a second payment of $19 million USD.

Speaker #2: Our AI peptide projects with Gannon Lee Pharmaceutical were also recognized as a Beijing key lab. These collaborations validate the platform's reproducibility and commercial landing across modalities and therapeutic areas.

Jian Ma: These collaborations validate the platform's reproducibility and commercial landing across modalities and therapeutic areas. As AI gets increasingly stronger at molecular design, we are increasingly aware that these predictions ultimately need validation in biological systems closer to the human body. Traditional cell and animal experiments still differ differently and significantly from a real human response. We want to build the next generation biological simulation platform to assess candidate efficacy and safety earlier and more accurately in vitro and reduce clinical translation risk. On that basis, we have invested and incubated three directions: virtual cells, organ-on-a-chip, and organoids, including in several cell biosystems and Signet Therapeutics, and thus successfully completed an angel round of tens of millions of RMB in H1, with participation from Shunwei Capital, Sequoia China, and Songhe Capital. Now moving to AI for science intelligence solutions.

Speaker #2: As AI becomes increasingly powerful at molecular design, we are increasingly aware that these predictions ultimately need validation in biological systems closer to the human body.

Speaker #2: Traditional cell and animal experiments still defer differently and significantly you know from a real human response. So we want to build the next generation biological simulation platform to assess candidate efficacy and safety earlier and more accurately in vitro and reduce clinical translation risk.

Speaker #2: On that basis, we have invested in and incubated three directions: virtual cells, organ-on-a-chip, and organoids, including several cellular biosystems and psychotherapeutics. We have also successfully completed an angel round of financing, raising tens of millions of RMB in the first half, with participation from Shenwei Capital, Sequoia China, and Songhe Capital.

Speaker #2: Now moving to AI for Science Intelligence Solutions. In the first half of 2026, this segment generated revenue of RMB 194 million, up 136.4% year-over-year. Behind that growth is an important industrial trend: whether in drug R&D or advanced materials, customers are increasingly accepting the use of robotic labs to generate high-quality experimental data and molecules.

Jian Ma: In the H1 2026, this segment generated revenue of RMB 194 million, up 136.4% year-over-year. Behind that growth is an important industrial trend, whether in drug R&D or advanced materials. Customers are increasingly accepting the use of robotic labs to generate high-quality experimental data and molecules, then feeding that data back into AI models to form a continuously iterating R&D loop. First, AI for Science intelligent robotic labs. That is Physical AI. In the first half, we delivered on a considerable number of flagship projects in China and overseas. Overseas, we continue to win recognition from leading customers. Our collaboration with Lilly has expanded from early solid dispensing to a compound inventory management system, and then further into High-Throughput Experimentation. Both the depth and system complexity keeps rising.

Speaker #2: Then, feeding that data back into AI models forms a continuously iterating R&D loop. First, AI for science intelligent robotic labs—that is physical AI.

Speaker #2: In the first half, we delivered on a considerable number of flagship projects in China and overseas. Overseas, we continue to win recognition from leading customers.

Speaker #2: Our collaboration with Lilly has expanded from early solid dispensing to a compound inventory management system, and then further into high-throughput experimentation. Both the depth and system complexity keep rising.

Speaker #2: And in terms of South Korea, the JW project in South Korea was fully delivered in April, and customer feedback has been very positive. In China, we've also got multiple projects, each worth more than 10 million RMB.

Jian Ma: In terms of South Korea, the JW project in South Korea was fully delivered in April, and customer feedback has been very positive. In China, we have also got multiple projects, each worth more than RMB 10 million. Intelligent synthesis workstations have been replicated across 13 customers. Our underlying capability is no longer limited to pharma. It has expanded into advanced materials such as perovskites, lithium batteries, and molecular sieves. This shows that the Physical AI plus agent foundation has strong cross-industry reusability potential. Now, the second part is AI for Science intelligence services. Commercialization accelerated in the H1 2026. The core driver is urgent demand from drug discovery customers to expand chemical space, the need to raise early discovery efficiency, and shorten project timelines. New order value grew rapidly in the first half, reflecting strong demand and continued customer recognition of our technology.

Speaker #2: Intelligent synthesis workstations have been replicated across 13 customers. Our underlying capability is no longer limited to pharma; it has expanded into advanced materials such as perovskites, lithium batteries, and molecular sieves.

Speaker #2: This shows that the Physical AI plus agent foundation has strong cross-industry reusability potential. Now, the second part is AI for science intelligence services. Commercialization accelerated in the first half of 2026.

Speaker #2: The core driver is urgent demand from drug discovery customers to expand chemical space, as well as the need to raise early discovery efficiency and shorten project pipelines.

Speaker #2: Timelines' new order value grew rapidly in the first half, reflecting strong demand and continued customer recognition of our technology. A key breakthrough was Vast, our virtual compound library, which won an order covering more than 20,000 molecules and opened a new growth space for us.

Jian Ma: A key breakthrough was VAST, our virtual compound library, which won an order covering more than 20,000 molecules and opened a new growth space for us. Behind that is a core early discovery pain point, how to explore vast chemical space faster and more efficiently and effectively, and find candidates that are synthesizable, screenable, and structurally diverse. The value of VAST is that it is not simply a compound list. It uses continuous feedback from automated experiments and proprietary data to help customers explore chemical space more efficiently and obtain higher quality, more diverse candidates, accelerating later screening and development. As one of the first companies to close the loop between AI computation and automated experimentation, we have turned large-scale chemical space exploration, previously hard to do efficiently, into a standardized deliverable product and service. This is a hard-to-replicate area.

Speaker #2: Behind that is a core early discovery—core early discovery pain points are how to explore vast chemical space faster and more efficiently and effectively, and find candidates that are synthesizable, screenable, and structurally diverse.

Speaker #2: The value of VAST is that it is not simply a compound list. It uses continuous feedback from automated experiments and proprietary data to help customers explore chemical space more efficiently and obtain higher-quality, more diverse candidates, accelerating later screening and development.

Speaker #2: As one of the first companies to close the loop between AI computation and automated experimentation, we have turned large-scale chemical space exploration—previously hard to do efficiently—into a standardized, deliverable product and service.

Speaker #2: This is a hard-to-replicate area. It has helped us basically define and create new demands, opening more growth space for AI for science intelligence services.

Jian Ma: It has helped us basically define and create new demands, opening more growth space for AI for Science intelligence services. This is definitely very important for us, and it has become one of our core competitive modes. As we mentioned, we actually foresee great growth potential based on this capability. We have also built and commercialized two solutions, Agentic Synthesis and Agentic HTE, enabling AI to make autonomous decisions across the full workflow. Now, think of Agentic Synthesis as a digital chemistry team driven by AI agents. A chemist only needs to describe the target molecule and experimental needs in natural language. The system can then complete an end-to-end loop. Target confirmation, raw material procurement, route design, experiment execution, testing, workup, purification, QC, and delivery. Our four core models have reached production-grade accuracy, and seven dedicated agents form a complete digital chemistry team.

Speaker #2: And this is definitely very important for us, and it has become one of our core competitive moats. As we mentioned, we actually foresee great growth potential based on this capability.

Speaker #2: We have also built and commercialized two solutions: Agentic Synthesis and Agentic HTE, enabling AI to make autonomous decisions across the full workflow. Now, think of Agentic Synthesis as a digital chemistry team driven by AI agents.

Speaker #2: A chemist only needs to describe the target molecule and experimental needs in natural language. The system can then complete an end-to-end loop: target confirmation, raw material procurement, route design, experiment execution, testing, workup, purification, QC, and delivery.

Speaker #2: Our four core models have reached production-grade accuracy, and seven dedicated agents form a complete digital chemistry team. The platform now has a capacity of about 8,000 synthesized molecules a month.

Jian Ma: The platform now has capacity of about 8,000 synthesized molecules a month, which is incredible. Success rates are above 95% for simple reactions and above 90% for complex reactions. This not only raises capacity, it also lets a single chemist move from developing reactions one by one to supervising experiments in batches. Agentic HTE is our autonomous high throughput experimentation solution. Traditional HTE condition optimization usually takes 3 to 4 weeks per iteration. Agentic HTE can shorten that to about 6 days. The core is not simply higher throughput. It is combining intent understanding, skill orchestration, long task governance, and human AI collaboration. The system can automatically break an experimental goal into executable tasks and track multi-day workflows.

Speaker #2: Which is incredible. Success rates are above 95% for simple reactions and above 90% for complex reactions. This not only raises capacity, it also lets a single chemist move from developing reactions one by one to supervising experiments in batches.

Speaker #2: Agentic HTE is our autonomous high-throughput experimentation solution. Traditional HTE condition optimization usually takes three to four weeks per iteration. Agentic HTE can shorten that to about six days.

Speaker #2: The core is not simply higher throughput. It is combining intent understanding, skill orchestration, long-task governance, and human-AI collaboration, so the system can automatically break an experimental goal into executable tasks and track multi-day workflows.

Speaker #2: So overall, from robotic labs to VAST, to agentic synthesis and agentic HTE, we now have a complete AI-for-science solution system—from data generation to molecular design to experimental execution.

Jian Ma: Overall, from robotic labs to VAST, to Agentic Synthesis and Agentic HTE, we now have a complete AI for Science solution system, from data generation to molecular design, to experimental execution. That also opens more space for subsequent growth. I will stop here. I will now invite Shuhao to discuss our layout in advanced materials and consumer health, and to introduce the XtalPi Science platform. Thank you, Dr. Ma. Now for our advanced materials and consumer health, let's talk about our AI autonomous R&D first. This core capability has been systematically validated in drug discovery, and also is now creating reusable cross-industry synergies and is accelerating into new industry scenarios such as advanced materials and consumer health. In other words, we are moving from deep focus on a single industry to platform-style replication. The general value of AI for Science infrastructure is starting to be released.

Speaker #2: That also opens more space for subsequent growth. I'll stop here. I'll now invite Shuhao to discuss our layout in advanced materials and consumer health, and to introduce the XtalPi Science to you.

Shuhao Wen: Thank you, Dr. Ma. Now for our advanced materials and consumer health, let's talk about our AI autonomous R&D first. This core capability has been systematically validated in drug discovery, and also is now creating reusable cross-industry synergies and is accelerating into new industry scenarios such as advanced materials and consumer health. In other words, we are moving from deep focus on a single industry to platform-style replication. The general value of AI for Science infrastructure is starting to be released.

Speaker #2: Dr. Ma: Now, for our advanced materials and consumer health, let's talk about our AI autonomous R&D first. This core capability has been systematically validated in drug discovery.

Speaker #2: And also is now creating reusable cross-industry synth synergies, and is accelerating into new industry scenarios such as advanced materials and consumer health.

Speaker #2: In other words, we are moving from a deep focus on a single industry to platform-style replication. The general value of AI for science infrastructure is starting to be realized. We also have a clear view on advanced materials—they are a core underlying carrier for new energy, high-end manufacturing, and electronics.

Shuhao Wen: We also have a clear view on advanced materials. They are a core underlying carrier for new energy, high-end manufacturing, and electronics. The long-term space is very large, and the growth certainty is relatively strong. We have formed a dedicated advanced materials team and migrated the design, synthesize, test, analyze loop already validated in drug R&D. We have now made substantive breakthroughs in several frontier materials. Specifically, we are now working with PV leader, Jinko Solar, to build an automated production line and tackle new materials for perovskite tandem cells, continuously improving photoelectric conversion efficiency and device stability. We are now venturing into polymer composites, furthering the materials business. In addition, XtalPi has built an AI and robotic lab-driven perovskite formulation platform. It iterates literature mining, dedicated databases, AI parameter recommendation, and automated experimental validation.

Speaker #2: The long-term space is very large, and the growth certainty is relatively strong. We have formed a dedicated advanced materials team and migrated the design–synthesize–test–analyze loop already validated in drug R&D.

Speaker #2: We have now made substantive breakthroughs in several frontier materials. Specifically, we're now working with PV leader Jinko Solar to build an automated production line and tackle new materials for perovskite tandem cells, continuously improving photoelectric conversion efficiency and device stability.

Speaker #2: We're now venturing into polymer composites, furthering the materials business. In addition, XtalPi has built an AI- and robotic lab-driven perovskite formulation platform. It iterates literature mining, dedicated databases, AI parameter recommendations, and automated experimental validation.

Speaker #2: Compared to traditional manual R&D, the overall cycle has been shortened from four to six years to one to six months—a significant order-of-magnitude jump in R&D efficiency.

Shuhao Wen: Versus traditional manual R&D, the overall cycle has been shortened from 4 to 6 years to 1 to 6 months, a significant order to magnitude jump in R&D efficiency. In consumer health, the Groland brand has completed zero to one validation. Its anti-hair loss scalp essence ranked first in a new Tmall category ranking. And during the 618, the store entered the top 3 new personal care stores. The product matrix covers the full wash and care process. The two molecules licensed to the brand have completed China's new cosmetic ingredient filing. A version that can be sold through all domestic channels is expected to launch by the end of 2026. This means Groland has successfully entered the large anti-hair loss category, opening more space for subsequent volume and brand growth.

Speaker #2: In Consumer Health, the Growland brand has completed a zero-to-one validation. Its anti-hair loss scalp essence ranked first in a new Tmall category ranking, and during the 6.18, the store entered the top three new personal care stores. The product matrix covers the full wash and care process.

Speaker #2: The two molecules licensed to the brand have completed China’s new cosmetic ingredient filing. A version that can be sold through all domestic channels is expected to launch by the end of 2026.

Speaker #2: This means Growland has successfully entered the large anti-hair loss category, opening more space for subsequent volume and brand growth. Now, if we combine the earlier parts, we have completed the critical step from internal capability building to validation in real business scenarios.

Shuhao Wen: Now, if we combine the earlier parts, we have completed the critical step from internal capability building to validation in real business scenarios. The next strategic upgrade is very clear. Further standardize, platformatize, and open the capabilities we have already validated. Driven by that vision, in July, we officially launched the XtalPi Science platform and the Genius Agents Suite. XtalPi Science is the world's first comprehensive AIfS platform, integrating large language models, scientific agents, and large-scale robotic experiments. We further standardize and platformatize the integrated research infrastructure accumulated and validated in real projects, gradually building scientific infrastructure for research organizations worldwide and lowering the barriers to advanced research capabilities. Genius Agents is the orchestration hub. It generates scientific hypotheses and specialist predictions in the digital world, then uses Physical AI for high-precision experimental validation in the physical world.

Speaker #2: The next strategic upgrade is very clear: further standardize, platformatize, and open the capabilities we have already validated. Driven by that vision, in July we officially launched the XtalPi Science Platform and the Genius Agent Suite.

Speaker #2: XtalPi science is the world's first comprehensive AI for science platform integrating large language models scientific agents and large scale robotic experiments. We further standardize and platformatize and you know the integrated research infrastructure accumulated and validated in real projects.

Speaker #2: Gradually building scientific infrastructure for research organizations worldwide, and lowering the barrier to advanced research capabilities. Genius Agent is the orchestration hub. It generates scientific hypotheses and specialist predictions in the digital world, then uses physical AI for high-precision experimental validation in the physical world.

Speaker #2: Result feedback is fed back into the system and drives the next round of iteration. This forms a complete loop of digital hypotheses, specialist prediction, physical validation, and data feedback.

Shuhao Wen: Results feed back into the system and drive the next round of iteration. This forms a complete loop of digital hypothesis, specialist prediction, physical validation, and data feedback. We have joined 26 partners to launch the Open Ecosystem Alliance for Science Intelligence. This shows that scientific intelligence infrastructure is moving from single company exploration to cross-industry, cross-institution co-building. Going forward, XtalPi Science also plans to use Science Token as a unified calling and metering mechanism for research resources, further improving resource allocation efficiency. I will stop here. I will invite Jeff to walk you through the company's financials. Thank you, Shuhao. Hi, everyone. I will now walk you through our H1 2026 financial performance. In H1, the company generated revenue of 394 million RMB compared with 517 million RMB in the same period of 2025, excluding a large pipeline BD deal. Revenue grew by 74% year over year.

Speaker #2: We have joined 26 partners to launch the Open Ecosystem Alliance for Science Intelligence. This shows that scientific intelligence infrastructure is moving from single company exploration to cross-industry, cross-institution co-building going forward. XtalPi Science also plans to use Science Token as a unified calling and metering mechanism for research resources, further improving resource allocation efficiency.

Speaker #2: I'll stop here. I'll invite Jeff to walk you through the company's financials. Thank you, Shuhao. Hi, everyone. I'll now walk you through our first half 2026 financial performance.

Feiran Zhou: Thank you, Shuhao. Hi, everyone. I will now walk you through our H1 2026 financial performance. In H1, the company generated revenue of 394 million RMB compared with 517 million RMB in the same period of 2025, excluding a large pipeline BD deal. Revenue grew by 74% year over year.

Speaker #2: In the first half, the company generated revenue of 394 million RMB compared with 517 million RMB in the same period of 2025.

Speaker #2: Excluding a large pipeline BD deal, revenue grew by 74% year-over-year for the same period. We continue to increase R&D investment, with R&D expenses up about 66% year-over-year. First half net loss was RMB 225 million, compared with net profit of RMB 76 million a year earlier. Adjusted net loss was RMB 106 million, compared with adjusted net profit of RMB 142 million a year earlier. The swing to loss mainly reflects the year-on-year decline in revenue and the 66% increase in R&D expenses.

Feiran Zhou: On the same period, we continued to increase R&D investment with R&D expenses up about 66% year over year. H1 net loss was 225 million RMB, compared with net profit of 76 million RMB a year earlier. Adjusted net loss was 106 million RMB, compared with adjusted net profit of 142 million RMB a year earlier. A swing to a loss mainly reflect the year-on-year decline in revenue and 66% increase in R&D expenses. The R&D increase was driven by continued investment in autonomous lab and Agentic system, as well as pipeline programs and multi-model technology platforms, which create greater space for our mid to long-term value increase. Excluding the large pipeline BD, organic revenue grew about 74% in H1.

Speaker #2: The R&D, you know, increase was driven by continual investment in autonomous lab and agentic systems, as well as pipeline programs and multimodal technology platforms, which create greater space for our near-term and long-term value increase.

Speaker #2: Excluding the large pipeline BD, organic revenue grew about 74% in the first half. Over the same period, the amount of adjusted loss, excluding large pipeline BD, was basically flat, while the loss ratio narrowed significantly by 58 percentage points. Operating quality continued to improve.

Feiran Zhou: Over the same period, the amount of adjusted loss excluding large pipeline BD was basically flat, while the loss ratio narrowed significantly by 58 percentage points. Operating quality continued to improve. First, drug discovery solutions. H1 revenue was 200 million RMB compared with 435 million a year earlier. The prior year period recognized $51 million USD upfront and a large pipeline BD, so the base was high. The collaboration is progressing well, and we have received a second contractual payment of $19 million USD. Excluding the large pipeline BD, drug discovery solutions revenue grew by 1% year over year. Now, what is worth noting is that we have actively advanced a strategic shift from services to assets by building proprietary innovative pipeline and multi-model platforms. We are accumulating high quality assets with strong clinical value and commercialization potential for mid to long-term growth. Now for AI for Science intelligent solutions.

Speaker #2: First Drug Discovery Solutions. First half revenue was RMB 200 million, compared with RMB 435 million a year earlier. The prior year period recognized $51 million upfront and a large pipeline BD.

Speaker #2: So the base was high. The collaboration is progressing well, and we have received a second contractual payment of $19 million. Excluding the large pipeline BD drug discovery solutions, revenue grew by 1% year over year.

Speaker #2: Now, what's worth noting is that we have actively advanced a strategic shift from services to assets by building proprietary, innovative pipeline and multimodal platforms.

Speaker #2: We are accumulating high quality assets with strong clinical value and commercialization potential for meter long-term growth. Now for AI for science intelligence solutions. First half revenue was 194 million RMB compared with 800 sorry compared with 82 million RMB a year earlier up 136% year over year.

Feiran Zhou: H1 revenue was 194 million RMB, compared with 82 million RMB a year earlier, up 136% year-over-year. Growth was driven by technology upgrades and high-quality delivery, which supported rapid new customer expansion and high repeat rates among existing customers. Both AI for Science intelligent robotic labs and AI for Science intelligent services maintain high growth, with the AIfS industry remaining upbeat. Our AI for Science commercialization has accelerated. As was mentioned by the Chairman and CEO, we actually saw new orders signed with great growth potential and momentum this year. Geographically speaking, in H1 2026, we continued to expand our global customer and flagship collaborations. Excluding the large pipeline licensing projects, revenue grew strongly year-over-year in every region. US market grew 14% and other overseas regions grew 108%.

Speaker #2: Growth was driven by technology upgrades and high-quality delivery, which supported rapid new customer expansion and a high repeat rate among existing customers. Both AI for Science Intelligent Robot Robotic Labs and AI for Science Intelligence Services maintained high growth, with the AI for Science industry remaining upbeat.

Speaker #2: Our AI for science commercialization has accelerated, and as was mentioned by the chairman and CEO, we actually saw new orders signed with great growth potential and momentum this year.

Speaker #2: Now, geographically speaking, in the first half of 2026, we continued to expand our global customer and flagship collaborations, excluding the large pipeline licensing projects.

Speaker #2: Revenue grew strongly year over year in every region. The US market grew 14%, and other overseas regions grew 108%. In drug discovery solutions, we landed multiple BD deals in China and overseas, including an AI drug discovery collaboration of more than $400 million with a well-known international pharmaceutical company, and several collaborations with innovative Chinese drug companies as well.

Feiran Zhou: In drug discovery solutions, we landed multiple BD deals in China and overseas, including an AI drug discovery collaboration of more than $400 million USD with a well-known international pharmaceutical company, and several collaborations with innovative Chinese drug companies as well. In AI for Science intelligence solutions, we are accelerating internationally. The robotic lab business has already delivered to leading overseas customers and generated repeat orders, including high-quality delivery and acceptance of the Eli Lilly projects and the 10 million-plus RMB JW projects in South Korea. By business mix, the share of AI for Science intelligence solutions rose sharply year-over-year from 16% to 49% of revenue as the AIfS industry remains upbeat. Physical AI robotic lab solutions are growing as industry consensus lands. The AI for Science intelligence services are matching new drug R&D demand with faster commercialization.

Speaker #2: In AI for science intelligence solutions, we are accelerating internationally. The robotic lab business has already delivered to leading overseas customers and generated repeat orders, including high-quality delivery and acceptance of the Eli Lilly projects and the 10 million-plus RMB JW projects in South Korea.

Speaker #2: By business, the share of AI for science intelligence solutions rose sharply year over year, from 16% to 49% of revenue, as the AI for science industry remains upbeat.

Speaker #2: Physical AI robotic lab solutions are growing as industry consensus lands. The AI-for-science intelligence services are matching new drug R&D demand with faster commercialization.

Speaker #2: Platform value is being realized faster, and revenue has grown in a step-change way. Operating expenses for the first half—R&D expenses were 368 million RMB, compared with 222 million RMB a year earlier, up 66%.

Feiran Zhou: Platform value is being realized faster and revenue has grown in a step change way. Operating expenses. H1 R&D expenses were 368 million RMB compared with 222 million RMB a year earlier, up 66%. This reflected continued investment in autonomous lab and Agentic systems, as well as pipeline programs and multi-model programs. General and administrative expenses were 212 million RMB compared with 200 million a year earlier, up 6%, reflecting higher infrastructure investment and higher employee benefit expenses to support business growth. Selling and marketing expenses were 48 million RMB compared with 40 million RMB a year earlier, up 19%, mainly due to higher employee benefit expenses as we continue to strengthen business development and market penetration. Cost of sales was 168 million RMB compared with 82 million a year earlier, up 105%, mainly because of larger business scale which drove higher fulfillment and delivery.

Speaker #2: This reflected continual investment in autonomous lab and agentic systems, as well as pipeline programs and multimodal programs. General and administrative expenses were ¥212 million RMB, compared with ¥200 million a year earlier, up 6%, reflecting higher infrastructure investment and higher employee benefit expenses to support business growth.

Speaker #2: Selling and marketing expenses were RMB 48 million, compared with RMB 40 million a year earlier, up 19%, mainly due to higher employee benefit expenses as we continue to strengthen business development and market penetration.

Speaker #2: Cost of sales was 168 million RMB, compared with 82 million a year earlier, up 105%, mainly because of larger, you know, business scale, which drove higher fulfillment and delivery.

Speaker #2: The company's financial position remains solid as of the first half. The cash balance, which is ¥8.67 billion RMB, is mainly due to proceeds from the convertible bonds issued in January 2026.

Feiran Zhou: The company's financial position remains solid. As of H1, the cash balance reached 8.67 billion RMB, mainly due to proceeds from the convertible bonds issued in January 2026. This ample reserve strongly supports continued R&D investment and helps consolidate our industry position in AI for Science. That is the main picture for H1 2026 finances. We will continue to drive higher quality long-term value release through AI for Science scale-up, global customer expansion, and platform and asset capability building. Thank you. Thank you, Dr. Wen, Dr. Ma, and Jeff for their remarks. That concludes our H1 2026 results presentation. We will now leave time for Q&A. I will first invite the conference secretary to explain how to ask questions. Hi, everyone. If you would like to ask a question on the English channel, please move on to the Chinese channel for your questions. Thank you. Hi, everyone.

Speaker #2: This ample reserve strongly supports continual R&D investment and helps consolidate our industry position in AI for science. That is the main picture for the first half of 2026 finances.

Speaker #2: We will continue to drive higher quality, long-term value release through AI for science, scale-up, global customer expansion, and platform and asset capability building. Thank you.

Speaker #2: Thank you, Dr. Wen, Dr. Ma, and Jeff, for their remarks. That concludes our first half 2026 results presentation. We will now leave time for Q&A.

Sun Choi Chung: Thank you, Dr. Wen, Dr. Ma, and Jeff for their remarks. That concludes our H1 2026 results presentation. We will now leave time for Q&A. I will first invite the conference secretary to explain how to ask questions.

Speaker #2: Our first invite is the conference secretary to explain how to ask questions. Hi, everyone. If you would like to ask a question on the English channel, please move to the Chinese channel.

Operator: Hi, everyone. If you would like to ask a question on the English channel, please move on to the Chinese channel for your questions. Thank you. Hi, everyone.

Speaker #2: Thank you for your questions. Hi everyone. Participants on the English channel who wish to ask a question, please move over to the Chinese channel.

Sun Choi Chung: Participants on the English channel who wish to ask a question, please move on over to the Chinese channel. Thank you. All right. The next question comes from our participant with a phone number ending with 0086. Please state your name and organization before you ask your question. Thank you. Hi, everyone. My name is Huang Yang from JPMorgan. My question is that the company used to focus on R&D services. In 2026, we already see that company is now gradually building proprietary pipelines. Actually, three are in clinical trial stages, and you have multiple that are preclinical, and some are in PCC, et cetera. From a mid to long-term strategy and resource allocation perspective, what is the core rationale for this shift, and how will R&D services and pipeline R&D be positioned, and how will they basically reinforce each other? Thank you. All right. Great.

Operator: Participants on the English channel who wish to ask a question, please move on over to the Chinese channel. Thank you. All right. The next question comes from our participant with a phone number ending with 0086. Please state your name and organization before you ask your question. Thank you.

Speaker #2: Thank you. All right, the next question comes from our participant with a phone number ending in 0086. Please state your name and organization before asking your question.

Speaker #2: Thank you. Hi everyone. My name is Huang Yang from JPM. My question is that the company used to focus on R&D services, and now in 2026 we already see that the company is gradually building proprietary pipelines. Actually, three are in clinical trial stages, and you have multiple that are preclinical and some that are in PCC, etc.

Yang Huang: Hi, everyone. My name is Huang Yang from JPMorgan. My question is that the company used to focus on R&D services. In 2026, we already see that company is now gradually building proprietary pipelines. Actually, three are in clinical trial stages, and you have multiple that are preclinical, and some are in PCC, et cetera. From a mid to long-term strategy and resource allocation perspective, what is the core rationale for this shift, and how will R&D services and pipeline R&D be positioned, and how will they basically reinforce each other? Thank you.

Speaker #2: So, from a mid to long-term strategy and resource allocation perspective, what is the core rationale for this shift, and how will R&D services and pipeline R&D be positioned? And how will they basically reinforce each other?

Speaker #2: Thank you. All right. Great. Thank you. This is a great Jen. Let me try to address your question. First of all regarding this shift on strategy these kind of proprietary pipeline or building such a proprietary pipeline is a natural extension of our capability boundary.

Jian Ma: All right. Great.

Yang Huang: Thank you. This is a great question. I am Ma Jian. Let me try to address your question. First of all, regarding this shift on strategy, these kind of proprietary pipeline or building such a proprietary pipeline is a natural extension of our capability boundary. Sorry, we are now putting the investor on mute because there were some noises generated. Please continue. Over the past few years, we have always been deeply involved in real industrial drug R&D and have accumulated broad capabilities from algorithms and experiments to program advancement. Moving from R&D services to proprietary pipelines is not starting a new business. It is further deepening of capabilities that we already have. Strategically speaking, proprietary pipelines have three kinds of value. First, they directly validate and release our platform capability in real drug discovery settings.

Jian Ma: Thank you. This is a great question. I am Ma Jian. Let me try to address your question. First of all, regarding this shift on strategy, these kind of proprietary pipeline or building such a proprietary pipeline is a natural extension of our capability boundary.

Speaker #2: Sorry. We are now putting the investor on mute because there was some noise generated. Please continue. So, over the past few years, we have—now, you know—we have always been deeply involved in real industrial drug R&D and have accumulated broad capabilities, from algorithms and experiments to program advancement.

Operator: Sorry, we are now putting the investor on mute because there were some noises generated. Please continue.

Jian Ma: Over the past few years, we have always been deeply involved in real industrial drug R&D and have accumulated broad capabilities from algorithms and experiments to program advancement. Moving from R&D services to proprietary pipelines is not starting a new business. It is further deepening of capabilities that we already have. Strategically speaking, proprietary pipelines have three kinds of value. First, they directly validate and release our platform capability in real drug discovery settings.

Speaker #2: So, moving from R&D services to proprietary pipelines is not starting a new business. It's a further deepening of capabilities that we already have. Now, strategically speaking, proprietary pipelines have three kinds of value.

Speaker #2: First, they directly validate and release our platform capability in real drug discovery settings. Second, the pharmaceutical know-how, experimental data, and project experience generated as pipeline advances feedback, you know, into our AI for science infrastructure.

Jian Ma: Second, the pharmaceutical know-how, experimental data, and project experience generated as pipeline events feedback into our AI for Science infrastructure, driving continuous iteration of models, data, and robotic experimental system. Third, commercially speaking, proprietary pipelines also can release asset value over time through upfront milestones and potential sales royalties. Doing pipelines and doing services are not in conflict. On the contrary, doing pipelines helps us deepen collaboration with leading pharma customers. The drug R&D experience accumulated as pipelines events helps us better understand those customers' needs and pain points. Looking back at the past decade of development, we are one of the pioneers that uses AI technology for drug discovery. We started by constructing this dry and wet lab experimental capability to the current level of intelligent services, and we have built this basic foundational data platform.

Speaker #2: Driving continuous iteration of models, data, and robotic experimental systems. Third, commercially speaking, you know, proprietary pipelines also can release asset value over time through upfront milestones and potential sales royalties.

Speaker #2: So, doing pipelines and doing services are not in conflict. On the contrary, doing pipelines helps us deepen collaboration with leading pharma customers. The drug R&D experience accumulated as pipelines advance helps us better understand those customers' needs and pain points.

Speaker #2: Looking back at the past decade of development you know we are one of the pioneers that uses AI technology for drug discovery we started by constructing this try and wet lab experimental capability to the current level of intelligent services and we have built this basic foundational data platform and I believe during this process we have proactively constructed our comprehensive R&D capability and this is very meaningful to support the industry development in particular drug development and we actually can generate high value assets especially for assets of scarcity and we have already see that since 2025 a lot of biotech pharma companies have done license out for international farmers based on their drug discovery capability as well and it has reached a huge valuation and high value growth and for the past six months we actually saw a lot of well-known large overseas pharmaceutical companies which I'm not going to bother with naming them for example you can just think of the top 20 big farmers actually have actually become more competitive and we definitely need to utilize AI capability to increase and raise the drug discovery capability.

Sun Choi Chung: I believe during this process, we have proactively constructed our comprehensive R&D capability, and this is very meaningful to support the industry development, in particular, drug development. We actually can generate high-value assets, especially for assets of scarcity. We have already seen that since 2025, a lot of biotech pharma companies have done license out for international pharmas based on their drug discovery capability as well, and it has reached a huge valuation and high-value growth. For the past six months, we actually saw a lot of well-known large overseas pharmaceutical companies, which I am not going to bother with naming them. For example, you can just think of the top 20 big pharmas actually become more competitive. We definitely need to utilize AI capability to increase and raise the drug discovery capability.

Jian Ma: I believe during this process, we have proactively constructed our comprehensive R&D capability, and this is very meaningful to support the industry development, in particular, drug development. We actually can generate high-value assets, especially for assets of scarcity. We have already seen that since 2025, a lot of biotech pharma companies have done license out for international pharmas based on their drug discovery capability as well, and it has reached a huge valuation and high-value growth. For the past six months, we actually saw a lot of well-known large overseas pharmaceutical companies, which I am not going to bother with naming them. For example, you can just think of the top 20 big pharmas actually become more competitive. We definitely need to utilize AI capability to increase and raise the drug discovery capability.

Speaker #2: At the same time, you know, the reason why we shifted our strategy is because of our infrastructure. We have accumulated industry know-how and platforming capabilities.

Sun Choi Chung: At the same time, the reason why we shift our strategy is because of our infrastructure. We have accumulated industry know-how and platforming capabilities. A lot of large pharmas really were eager for that kind of infrastructure, and that is why we have just naturally and organically acquired such an opportunity by shifting to a dual driver business model. Thank you. The next question comes from an investor with a phone number ending with 7592. Please state your name and organization before you ask your question. Hi. Dr. Ma, Dr. Wen, thank you so much for taking my question. I am from CITIC Securities. My name is Tan Rushen. Please, can you introduce the positioning and your business model of XtalPi Science in the company's overall strategy? At the same time, what stage of progress has been achieved so far? All right. Thank you for your question.

Jian Ma: At the same time, the reason why we shift our strategy is because of our infrastructure. We have accumulated industry know-how and platforming capabilities. A lot of large pharmas really were eager for that kind of infrastructure, and that is why we have just naturally and organically acquired such an opportunity by shifting to a dual driver business model.

Speaker #2: A lot of large farmers really were eager for that kind of infrastructure, and that is why we have just naturally and organically acquired such an opportunity by shifting to a dual, you know, driver business model.

Speaker #2: Thank you. The next question comes from an investor with a phone number ending in 7592. Please state your name and organization before you ask your question.

Operator: Thank you. The next question comes from an investor with a phone number ending with 7592. Please state your name and organization before you ask your question.

Speaker #2: Hi, Dr. Ma, Dr. Wen, thank you so much for taking my question. I am from Citig Computer. My name is Han Ruchen. Could you please introduce the positioning and business model of XtalPi Science within the company's overall strategy?

Zhu Chen: Hi. Dr. Ma, Dr. Wen, thank you so much for taking my question. I am from CITIC Securities. My name is Chen Zhu. Please, can you introduce the positioning and your business model of XtalPi Science in the company's overall strategy? At the same time, what stage of progress has been achieved so far?

Speaker #2: At the same time, you know what stage of progress has been achieved so far. All right, thank you for your question. Let me try to address this question.

Shuhao Wen: All right. Thank you for your question.

Shuhao Wen: Let me try to address this question. XtalPi Science is positioned as the platform vehicle for the company's AI for Science capabilities, not as a separate new business line. It is very important and critical for the company's development. Over the past decade or 10 plus years, we have accumulated scientific AI models, specialist data, R&D tools, chemistry capability, and robotic labs. With Genius Agents at the core, XtalPi Science strings these capabilities together so we can keep advancing a complete R&D task around the customer's goal. The core value is to make our existing capabilities more standardized, molecular, and reusable, so customers can flexibly call models, data, compute, and experimental resources through one unified entry point. That helps us go deeper with each customer, and also copy validated capabilities across more customers and R&D scenarios, raising delivery efficiency and scale potential.

Speaker #2: XtalPi Science is positioned as the platform vehicle for the company's AI-for-science capabilities, not as a separate new business line. It is very important and critical for the company's development.

Speaker #2: Over the past, you know, decade or 10-plus years, we have accumulated scientific AI models, specialist data, R&D tools, chemistry capability, and robotic labs.

Speaker #2: With Genius Agent at the core, XtalPi Science strings these capabilities together so we can keep advancing a complete R&D task around the customer's goals.

Speaker #2: The core value is to make our existing capabilities more standardized, modular, and reusable, so customers can flexibly call models, data, compute, and experimental resources through one unified entry point. That helps us go deeper with each customer and also copy validated capabilities across more customers and R&D scenarios, raising delivery efficiency and scale potential.

Speaker #2: Now that is part of the core value in the business model. Going forward, we can charge for platform and agent usage, model and data costs, compute, and automated experiment execution.

Shuhao Wen: Now, that is part of the core value. On the business model, going forward, we can charge for platform and agent usage, model and data costs, and compute an automated experiment execution. We can also provide project-based R&D services and co-building enterprise R&D systems. Our Science Token is mainly used for unified calling and metering of different research resources on the platform. Now looking at near-term progress, XtalPi Science officially launched in July 2026. We plan to invite selected customers to a trial in September 2026 and gradually open for platform capabilities. As you mentioned, this is a very strategic part of the company, and we believe in the future of AI for Science. There will be companies that are worth over 100 billion USD in industry to be born. Based on our introduction, globally speaking, we are well-positioned.

Speaker #2: We can also provide project-based R&D services and co-building enterprise R&D systems. Our science token is mainly used, you know, for unified calling and maturing of different research resources on the platform.

Speaker #2: Now, looking at near-term progress, XtalPi Science officially launched in July 2026. We plan to invite selected customers to a trial in September 2026 and gradually open up platform capabilities.

Speaker #2: As you mentioned, you know this is a very strategic part of the company, and we believe in the future of AI for science. There will be companies that are worth over $100 billion in this industry to be born, and based on our introduction, globally speaking, we are well positioned.

Speaker #2: As a publicly listed company, our goal and strategic positioning are to become one of the top players in the world. Thank you for the answer.

Sun Choi Chung: As a listed, publicly listed company, our goal and strategic positioning are to become one of the top players in the world. Thank you for the answer. I don't have any further questions. Thank you. All right. The next question comes from an investor with a phone number ending 5757. Hi, everyone. My name is Chen Bingyu from Jefferies. First of all, we would like to congratulate the company for the excellent results from the H1 2026. We would like to understand the key drivers for this growth, and can this growth continue for the full year? Was the stronger AIDD industry backdrop the main driver of higher related orders? Hi, my name is Jian Ma. As we mentioned, in the H1 2026, AI for Science Intelligence Solutions generated revenue of 194 million RMB, up 136.4% year-over-year.

Sun Choi Chung: As a listed, publicly listed company, our goal and strategic positioning are to become one of the top players in the world.

Zhu Chen: Thank you for the answer. I don't have any further questions. Thank you.

Speaker #2: I don't have any further questions. Thank you. All right, the next question comes from an investor with a phone number ending in 5757. Hi, everyone.

Operator: All right. The next question comes from an investor with a phone number ending 5757.

Chen Bingyu: Hi, everyone. My name is Chen Bingyu from Jefferies. First of all, we would like to congratulate the company for the excellent results from the H1 2026. We would like to understand the key drivers for this growth, and can this growth continue for the full year? Was the stronger AIDD industry backdrop the main driver of higher related orders?

Speaker #2: My name is Chen Bingyu from Jefferies. First of all, we would like to congratulate the company on the excellent results from the first half.

Speaker #2: For 2026, we would like to understand the key drivers for this growth. Can this growth continue for the full year? Was a stronger AIDD industry backdrop the main driver of higher related orders?

Speaker #2: Hi, my name is Ma Jian. As we mentioned in the first half of 2026, AI for science intelligence solutions generated revenue of RMB 194 million, up 136.4% year over year. Both intelligent robotic labs and intelligent services remained, you know, high growth.

Jian Ma: Hi, my name is Jian Ma. As we mentioned, in the H1 2026, AI for Science Intelligence Solutions generated revenue of 194 million RMB, up 136.4% year-over-year.

Jian Ma: Both intelligent robotic labs and intelligence services remained high growth. The core driver is that customer demand is upgrading from using a single model or tool to an R&D closed loop that combines high-quality data generation, AI model application, and experimental validation. In July, from a well-known conference, the government also mentioned that the country is moving towards a digital future, and that basically generated a lot of opportunities for different industries, particularly biopharmaceuticals and also the materials development featuring a critical important value-driven strategic application. From the company's perspective, we have accumulated a lot of experience and industry know-how thanks to the past decade of development, in terms of our algorithm, of the digital world, our professional-driven models, and our intelligent autonomous experimental platform. As a result, our capability is well-positioned to benefit from the external world.

Speaker #2: The core driver is that customer demand is upgrading from using a single model or tool to an R&D closed loop that combines high-quality data generation, AI model application, and experimental validation.

Speaker #2: Now, also in July, you know, from a well-known conference, the government also has mentioned that the company—I mean, the country—is moving towards a digital future, and that basically generates a lot of opportunities for different industries, particularly, you know, biopharmaceuticals and also materials development featuring critically important, value-driven, strategic applications.

Speaker #2: So from, you know, the company's perspective, we have accumulated a lot of experience and industry know-how thanks to the past decade of development. You know, in terms of our algorithms of the digital world, our professionally driven models, and our intelligent autonomous experimental platform. As a result, our capability is well positioned to benefit from the external world.

Speaker #2: With that said such best match of capabilities to market demand there is no reason for us to give up the opportunity and so the next step we need to increase the development of our intelligent robotic labs and that is why for the first half of the company you can already see that we actually have some finished a lot of successful collaboration with well-known players and a lot of downstream customers are now accelerating a new R&D model robotic labs built the data foundation and high quality data then feedbacked into AI models you know we completed delivery and acceptance of LLE's compound management and HT projects the JW project in South Korea completed overall delivery domestically intelligent synthesis workstations have been replicated across 13 customers.

Jian Ma: With that said, such best match of capabilities to market demand, there is no reason for us to give up the opportunity. The next step, we need to increase the development of our intelligent robotic labs. That is why for H1, you can already see that we actually have finished a lot of successful collaborations with well-known players. A lot of downstream customers are now accelerating a new R&D model. Robotic labs build the data foundation and high-quality data, then feed back into AI models. We completed delivery and acceptance of Eli Lilly's compound management and HTE projects. The JW project in South Korea completed overall delivery. Domestically, intelligent synthesis workstations have been replicated across 13 customers. On intelligent services, growth mainly came from our drug R&D demand to expand chemical space and shorten timelines.

Speaker #2: On Intelligent Services, you know growth mainly came from our drug R&D demand to expand chemical space and shorten timelines. Also, as we mentioned earlier, the vast virtual compound library went in order covering more than 20,000 molecules.

Jian Ma: Also, as we mentioned earlier, the VAST virtual compound library won an order covering more than 20,000 molecules. Agentic Synthesis and Agentic HTE have both been commercialized. Agentic HTE shortens a traditional HTE cycle from three to four weeks to about six days. These capabilities have been expanded into advanced materials, bringing incremental demand beyond traditional AIDD. These are a huge success. As you can see from our previous presentation and sharing, we have witnessed great development milestones that have gone beyond biopharmaceutical sector into, for example, new material development. A stronger AIDD industry backdrop is an important external tailwind, but it is not the main driver of order growth. The more important support is our long-accumulated Physical AI platform capability, plus project delivery, product replication, and cross-industry expansion. The industry backdrop creates a better market environment.

Speaker #2: Agentic synthesis and agentic HTE have both been commercialized. Agentic HTE shortens a traditional HTE cycle from three to four weeks to about six days.

Speaker #2: These capabilities have been expanded into advanced materials, bringing incremental demand beyond traditional AIDD. So these are huge successes, and as you can see from our previous presentations and sharing, we have witnessed great development milestones that have gone beyond the biopharmaceutical sector into, for example, new material development.

Speaker #2: Now, a stronger AIDD industry backdrop is an important external tailwind, but it is not the main driver of order growth. The more important support is our long-accumulated fiscal AI platform capability, plus project delivery, product replication, and cross-industry expansion.

Speaker #2: The industry backdrop creates a better market environment. Our own technology and engineering capability determine whether demand can actually convert into orders and revenue. Now, as we mentioned, looking into the full year of 2026, AI for science intelligence solutions have some project-based characteristics, so revenue recognition is affected by delivery and acceptance timelines. Based on the current project backdrop and the delivery schedule, second-half deliveries are expected to be more concentrated.

Jian Ma: Our own technology and engineering capability determine whether demand can actually convert into orders and revenue. As we mentioned, looking into the full year of 2026, AI for Science intelligence solutions has some project-based characteristics, so revenue recognition is affected by delivery and acceptance timeline. Based on the current project backdrop and the delivery schedule, H2 deliveries are expected to be more concentrated as projects complete acceptance and as new solutions and application scenarios expand further. If projects proceed as planned, we still expect a relatively solid full-year growth level. That is basically my answer for your question. Thank you. Thank you, Dr. Ma, for your answer. Thank you. The next question comes from an investor with a telephone number ending 9361. Hi, my name is Ying Ying from CSE Computer.

Speaker #2: As projects complete acceptance, and as new solutions and application scenarios expand further, if projects proceed as planned, we still expect a relatively solid full-year growth level.

Speaker #2: And so that's basically my answer to your question. Thank you. Thank you, Dr. Ma, for your answer. Thank you. The next question comes from an investor with telephone number ending in 9361.

Chen Bingyu: Thank you, Dr. Ma, for your answer. Thank you.

Operator: The next question comes from an investor with a telephone number ending 9361.

Speaker #2: Hi, my name is Ying Ying from CSC Computer. My question is that the company has built a four-layer technology foundation. Versus large model companies such as Anthropic and OpenAI, where is the core differentiation of the company, and how will you continue to iterate and stay ahead?

Ying Ying: Hi, my name is Ying Ying from CSC Computer.

Ying Ying: My question is that the company has built a four-layer technology foundation versus large model companies such as Anthropic and OpenAI. Where is the core differentiation of the company, and how will you continue to iterate and stay ahead? Thank you for the question. Now, several points. First of all, we would like to highlight several advantages of ours. The first one is that different types of companies industry each have their own strengths, but XtalPi's positioning and path are fundamentally different from a general large model companies. Our differentiation is that we deeply couple vertical domain models, high-quality proprietary data, robotic experimental capability, and real drug R&D scenarios to form a more complete set of AI for science infrastructure. Compared with large model companies, our models are vertical models built around specific drug R&D problems.

Speaker #2: Thank you for the question. Now, several points. First of all, we would like to highlight several advantages of ours. The first one is that different types of companies in the industry each have their own strengths, but XtalPi's positioning and path are fundamentally different from general large model companies.

Shuhao Wen: Thank you for the question. Now, several points. First of all, we would like to highlight several advantages of ours. The first one is that different types of companies industry each have their own strengths, but XtalPi's positioning and path are fundamentally different from a general large model companies. Our differentiation is that we deeply couple vertical domain models, high-quality proprietary data, robotic experimental capability, and real drug R&D scenarios to form a more complete set of AI for science infrastructure. Compared with large model companies, our models are vertical models built around specific drug R&D problems.

Speaker #2: Our differentiation is that we deeply couple vertical domain models, high-quality proprietary data, robotic experimental capability, and real drug R&D scenarios to form a more complete set of AI-for-science infrastructure.

Speaker #2: Compared with large model companies, our models are vertical models built around specific truck R&D problems. At the same time, we can leverage China's supply chain advantage, which will be very competitive in the market.

Shuhao Wen: At the same time, we can leverage China's supply chain advantage, which will be very competitive in the market. If you look at more important side of the story is that the hard part of drug R&D has never been model prediction alone. It is also how to obtain high-quality experimental data, how to understand real R&D scenarios, and how to get models results into actual project decisions. Just to summarize, we have long accumulation of our own wet lab system, industry proprietary data, and a large body of real project experience. This form the company's core throat. I think these are our core advantages, both our hardware and our industry know-how and our software. Now, another advantage is our AI for science solution system, which are based on real data and real experiments and industry know-how.

Speaker #2: If you look at the more important side of the story, the hard part of drug R&D has never been model prediction alone. It is also about how to obtain high-quality experimental data, how to understand real R&D scenarios, and how to get models' results into actual project decisions.

Speaker #2: So just to, you know, summarize, we have a long accumulation of our own wet lab system, industry proprietary data, and a large body of real project experience.

Speaker #2: This forms the company's core throat. And I think these are our core advantages—both our hardware and our industry know-how and our software. Now, another advantage is our AI for Science solution system.

Speaker #2: Which are based on real data, real experiments, and industry know-how. As we mentioned, if you look at our comprehensive pipeline and programs, and also the successful projects that we have delivered to important clients such as Theion Enterprises in China, we have already validated our capabilities, and we keep iterating our capability together with the customer.

Shuhao Wen: As we mentioned, if you look at our comprehensive pipeline and programs, and also the success projects that we successfully deliver to important clients such as Sanyou Corporation Limited in China, we have already validated our capabilities, and we keep iterating our capability together with the customer. Our models are built around specific real-world challenges in drug discovery, things built in the real-world practice. This kind of precision and reliability is not easy to replicate. With that said, we are confident that we can continue to iterate our models that are vertical specific and built from real data and are read and valuable data as well. Going forward, we believe that we can continue to iterate our model, and we also see that a lot of large language model companies, especially in China, have adopted an open source form.

Speaker #2: And our models are built around specific real world challenges in drug discovery things built by you know in the real world practice. And this kind of precision and reliability is not easy to replicate with that said we are confident that we can continue to iterate our models that are vertical specific and built from real data and read and valuable data as well.

Speaker #2: Going forward we believe that we can continue to iterate our model and we also see that a lot of large language model companies have adopted you know especially in China have adopted an open source form and as you mentioned in your question there are several companies international large model companies that are very competitive and building on the foundation of continuously evolving open source large models we're very confident that in this industry we can remain a very competitive position.

Shuhao Wen: As you mentioned in your question, there are several companies, international large model companies, that are very competitive and building on the foundation of continuously evolving open source large models. We are very confident that in this industry, we can remain a very competitive position. Thank you, Dr. Wen. Thank you so much for your answer. We will continue to follow up, and we wish you the best, especially in AI for science. Thank you. Now, the next question comes from our investor with a phone number ending 8627. Please go ahead. Hi, my name is Cyrus from Deutsche Bank. I have two questions here. The first one is about the gross profit margin changes for the H1 2026. What were the key drivers for that? Another question is your R&D expenses which rose quite sharply year-over-year in the H1.

Speaker #2: Thank you, Dr. Wen. Thank you so much for your answer. We'll continue to follow up, and we wish you the best, especially in AI for science.

Ying Ying: Thank you, Dr. Wen. Thank you so much for your answer. We will continue to follow up, and we wish you the best, especially in AI for science. Thank you.

Speaker #2: Thank you. Now, the next question comes from our investor with a phone number ending in 8627. Please go ahead. Hi, my name is Cyrus from DB.

Operator: Now, the next question comes from our investor with a phone number ending 8627. Please go ahead.

[Analyst] (Deutsche Bank): Hi, my name is Cyrus from Deutsche Bank. I have two questions here. The first one is about the gross profit margin changes for the H1 2026. What were the key drivers for that? Another question is your R&D expenses which rose quite sharply year-over-year in the H1.

Speaker #2: I have two questions. The first one is about the, you know, gross profit margin changes for the first half of 2026. What would the key drivers be for that?

Speaker #2: Another question is about your R&D expenses, which rose quite sharply year over year in the first half. So my question is the key directions of R&D investment and the future investment pace. And lastly, with increasing R&D expenses and GP margin changes, what's your profit—what's your expected timeline for your mid- to long-term profitability?

[Analyst] (Deutsche Bank): My question is the key directions of R&D investment and the future investment pace. Lastly, with increasing R&D expenses and GP margin changes, what is your expected timeline for your mid to long term profitability and profit realization? Thank you. Let me try to address this question. First of all, regarding the R&D investment, as was mentioned by Dr. Wen and Dr. Ma, our industry is growing very, very fast. This period of R&D investment is concentrated on several directions. Because AI for Science is still in fast development stage, in many application scenarios, we have already seen AI bringing real changes and improvement in R&D efficiency and decision quality. We actually see larger room to expand. Our investment will focus on, first, continuing to raise model and algorithm capability, including vertical models for different drug modalities and key R&D steps.

Speaker #2: And profit realization. Thank you. Let me try to address this question. First of all, regarding the R&D investment, as was mentioned by Dr. Wen and Dr. Ma, you know our industry is growing very, very fast, and this period of R&D investment is concentrated on several directions.

Feiran Zhou: Let me try to address this question. First of all, regarding the R&D investment, as was mentioned by Dr. Wen and Dr. Ma, our industry is growing very, very fast. This period of R&D investment is concentrated on several directions. Because AI for Science is still in fast development stage, in many application scenarios, we have already seen AI bringing real changes and improvement in R&D efficiency and decision quality. We actually see larger room to expand. Our investment will focus on, first, continuing to raise model and algorithm capability, including vertical models for different drug modalities and key R&D steps.

Speaker #2: And because AI for science is still in fast development stage and in many application scenarios we have already seen AI brings you know bringing real changes and improvement in R&D efficiency and decision quality and we actually see larger room to expand and so our investment will focus on first continuing to raise model and algorithm capability including vertical models for different drug modalities and key R&D steps second strengthening robotic experimental platforms to raise high quality data generation and experimental validation efficiency third advancing proprietary pipelines further validating and accumulating platform capability through real projects.

Feiran Zhou: Second, strengthening robotic experimental platforms to raise high-quality data generation and experimental validation efficiency. Third, advancing proprietary pipelines, further validating and accumulating platform capability through real projects. Fourth, continuing to improve the AI for Science infrastructure platform so it can better serve external customers and internal pipelines. On investment pace, we will keep R&D investment prudent and focused. We will not simply cut key R&D spend for short-term profit. We currently have a relatively ample cash reserve around RMB 8.67 billion at period end, which can support continued investment in algorithm, automated experiments, and drug development. Over mid to long term, as AI for Science intelligent solutions scale, as proprietary pipeline asset value is gradually released, and as platform capability raises delivery efficiency, profitability should gradually improve.

Speaker #2: Fourth, continuing to improve the AI for science infrastructure platform so it can better serve external customers and internal pipelines. Our investment pace will keep R&D investment prudent and focused; we will not simply cut key R&D spend for short-term profit. We currently have a relatively ample cash reserve, around RMB 8.67 billion at period end, which can support continued investment in algorithms, automated experiments, and drug development.

Speaker #2: Over the mid to long term, as AI-for-science intelligent solutions scale, proprietary pipeline asset value is gradually released, and as platform capability raises delivery efficiency, profitability should gradually improve.

Speaker #2: While keeping a long-term investment, we will continue to watch operating efficiency and the pace of commercial realization, and work to balance technology leadership, revenue growth, and improving profitability.

Feiran Zhou: While keeping long-term investment, we will continue to watch operating efficiency and the pace of commercial realization and work to balance technology leadership, revenue growth, and improving profitability. Most importantly, we want to draw your attention to the fact that, first of all, we see continuous interaction from the customers. Secondly, we also have seen the infrastructure growth in the process. Also over 70% of our customers are existing customers or regular customers, which says a lot about the customer recognition to the company's capability. Overall, we are still very confident in the profitability of these two businesses. Some of the aforementioned impacts are temporary or periodic. As we continue to development with revenue, including the ramp-up in acceptance and settlement, we believe the alignment between the revenue and cost will gradually improve.

Speaker #2: And most importantly, we want to draw your attention to the fact that, first of all, we see continuous interaction from the customers. Secondly, we have also seen, you know, infrastructure growth in the process. Also, over 70% of our customers are existing or regular customers, which says a lot about the customer recognition of the company's capability.

Speaker #2: So overall, we are still very confident in the profitability of these two businesses, so some of the aforementioned impacts are temporary or periodic as we continue development with revenue, including the ramp-up in acceptance and settlement.

Speaker #2: We believe the alignment between the revenue and cost will gradually improve. The efficiency of resource utilization and the benefit of economies of scale will also, you know, be further released. We expect the improvement in profitability to be backed up in stages in the short term.

Feiran Zhou: The efficiency of resource utilization and the benefit of economies of scale will also be further released. We expect the improvement in profitability to be backed up in stages in the short term. We will monitor the project settlement and growth profit margin development. The company also will support the expansion of our business and improvement in platform interaction efficiency, plus the value of long-term pipeline development. We believe that in the long term, we will continue to enhance our operating quality and efficiency. Thank you. The next question come from our investor with the phone number ending 4826. Please go ahead. Hi, my name is Jung Yu from Guolian Minsheng Healthcare. I have two questions. The company has laid out different vertical models around different modalities. What technical synergies can these models form? Based on platform capability and market opportunity, which directions will future pipeline layout focus on?

Speaker #2: You know, we'll monitor the project settlement and gross profit margin development, and the company, you know, also will support the expansion of our business and improvement in platform, you know, interaction efficiency, plus the value of long-term, you know, pipeline development.

Speaker #2: We believe that in the long term, we'll continue to, you know, enhance our operating quality and efficiency. Thank you. The next question comes from our investor with a phone number ending in 4826.

Operator: The next question come from our investor with the phone number ending 4826. Please go ahead.

Speaker #2: Please go ahead. Hi, my name is Zheng Yu from Guolian Mingsheng Healthcare. I have two questions. Now that the company has laid out different vertical models around different model modalities, what technical synergies can these models form?

Yu Jun: Hi, my name is Jun Yu from Guolian Minsheng Healthcare. I have two questions. The company has laid out different vertical models around different modalities. What technical synergies can these models form? Based on platform capability and market opportunity, which directions will future pipeline layout focus on?

Speaker #2: Based on platform capability and market opportunity, which directions will the future pipeline layout focus on? So, based on your current platform capabilities and the analysis of market opportunities, can the company also share their outlook for the future?

Jung Yu: Based on your current platform capabilities and based on the analysis of the market opportunities, can the company also share their outlook for the future? Thank you, Jung Yu, for your question. I am Jian Ma. Let me try to address this question. We basically built vertical models around different drug modalities based on a long-term understanding of AI drug discovery scenarios. Drug R&D problems are very complex. Data forms, R&D paths, and key decision points differ greatly across modalities. It is hard for one general model to solve everything. We built proprietary models closer to specific scenarios for different modalities, and key R&D steps. That does not mean that the models are siloed. Many underlying capabilities can be shared and coordinated. For example, algorithm frameworks, data collection and processing methods, and experimental validation systems.

Yu Jun: Based on your current platform capabilities and based on the analysis of the market opportunities, can the company also share their outlook for the future?

Speaker #2: Thank you, Zheng Yu, for your question. I'm Ma Jian. Let me try to address this question. Now we basically built vertical models around different drug modalities based on long-term understanding of AI drug discovery scenarios.

Jian Ma: Thank you, Jun Yu, for your question. I am Jian Ma. Let me try to address this question. We basically built vertical models around different drug modalities based on a long-term understanding of AI drug discovery scenarios. Drug R&D problems are very complex. Data forms, R&D paths, and key decision points differ greatly across modalities. It is hard for one general model to solve everything. We built proprietary models closer to specific scenarios for different modalities, and key R&D steps. That does not mean that the models are siloed. Many underlying capabilities can be shared and coordinated. For example, algorithm frameworks, data collection and processing methods, and experimental validation systems.

Speaker #2: Drug R&D problems are very complex. Data forms, R&D paths, and key decision points differ greatly across modalities. It is hard for one general model to solve everything.

Speaker #2: So, we built proprietary models tailored to specific scenarios for different modalities and key R&D steps. That does not mean, however, that the models are siloed.

Speaker #2: Many underlying capabilities can be shared and coordinated. For example, algorithm frameworks, data collection and processing methods, and experimental validation systems. So, as we mentioned, we need to build models closer to different scenarios; that is the rationale.

Jian Ma: As we mentioned, we need to build models closer to different scenarios. That is the rationale. Different models within the whole R&D systems, they are not separated. As we mentioned, a lot of the underlying capabilities such as algorithm, frameworks, data collection and processing method, and experimental validation systems are shared. That is why we built agents on top of the models, to understand, break down, and dispatch more complex R&D tasks in a unified way. That is why, whether they are small molecules or large ones, the process of discovering molecules are the same. That is why, with the platform, we will be able to achieve synergy. Future AI drug discovery is likely not one model doing all the work, but multiple vertical models and agent systems working together on complex tasks. That is the direction our technology system continues to evolve.

Speaker #2: Now, different models within the whole R&D system are not separated. And, you know, as we mentioned, a lot of the underlying capabilities—such as algorithm frameworks, data collection and processing methods, and experimental unit validation systems—are shared. So that's why we built agents on top of the models, you know, to understand, break down, and dispatch more complex R&D tasks in a unified way.

Speaker #2: So that's why, you know, whether there are small molecules or large ones, the process of discovering molecules is the same. So that's why, you know, with the platform, we will be able to achieve synergy.

Speaker #2: Future AI drug discovery is likely not one model doing all the work, but multiple vertical models and agent systems working together on complex tasks.

Speaker #2: That is the direction our technology system continues to evolve. So based on this capability accumulation, future pipeline layout will focus on two types of direction.

Jian Ma: Based on this capability accumulation, future pipeline layout will focus on two types of direction. First, drug modalities and R&D steps that highly match our vertical models and experimental platforms, and that can show the advantage of AI enablement. Second, directions with clear clinical need, large market space, and a chance to form differentiated data and asset value, such as autoimmune and CNS. Overall, we will not simply chase pipeline counts. We will put more weight on fit with platform capability, advancement efficiency, and future BD and commercialization potential. That will be the directions that we are going towards. Because when we look at our models, we have a strategy for program optimization for the entire process as well. This is very, very important for us to optimize. That is why, as we mentioned, we are going to focus on these two directions for future development pipeline layout.

Speaker #2: First, drug modalities and R&D steps that highly match our vertical models and experimental platforms, and that can show the advantage of AI enablement. Second, directions with clear clinical need, large market space, and a chance to form differentiated data and asset value, such as autoimmune and CNS.

Speaker #2: Overall, we will not simply chase pipeline counts. We will put more weight on fit with platform capability, advancement efficiency, and future BD and commercialization potential.

Speaker #2: That will be the direction that we are going toward. Because when we look at our models, you know, we have a strategy for program optimization for the entire process as well.

Speaker #2: You know, this is very, very important for us to optimize. And that is why, as we mentioned, we're going to focus on these two directions of future development pipeline layout.

Speaker #2: Thank you, Dr. Ma. Can I ask a follow-up question? As you mentioned in your answer, if I understand this correctly, basically for large molecules you need to, you know, look at the affinity capability. Compared to small molecules, maybe large molecules have a better chance for some optimizations—it's got better potential, you know, as a drug.

Jung Yu: Thank you, Dr. Ma. Can I ask a follow-up question? As you mentioned in your answer, if I understand this correctly, basically for large molecules, you need to look at the affinity capability compared to small molecule. Maybe large molecules have a better chance for some optimizations. It has got better potential as a drug. It also posts special kind of challenge as well. In the coming years, when you look at large molecule pipeline preparation, what are you going to start with and how are you going to develop? Is it going to be based on the type of iterative approach built upon existing antibodies, for example, antibody engineering? Or rather, will you tackle more challenging tasks? Well, this is actually a very good question. When you look at XtalPi, as you know that a lot of pharmaceutical companies. Sorry, we cannot hear the management very clearly.

Yu Jun: Thank you, Dr. Ma. Can I ask a follow-up question? As you mentioned in your answer, if I understand this correctly, basically for large molecules, you need to look at the affinity capability compared to small molecule. Maybe large molecules have a better chance for some optimizations. It has got better potential as a drug. It also posts special kind of challenge as well. In the coming years, when you look at large molecule pipeline preparation, what are you going to start with and how are you going to develop? Is it going to be based on the type of iterative approach built upon existing antibodies, for example, antibody engineering? Or rather, will you tackle more challenging tasks?

Speaker #2: So, also, it poses a special kind of challenge as well. So in the coming years, when you look at the large molecule pipeline preparation, what are you going to start with, and how are you going to develop?

Speaker #2: Is it going to be based on the, you know, type of iterative approach built upon existing antibodies—for example, antibody engineering—or rather will you tackle, you know, more challenging tasks?

Speaker #2: Well, this is actually a very good question. When you look at XtalPi, as you know, a lot of pharmaceutical companies—sorry, we cannot hear the management very clearly.

Jian Ma: Well, this is actually a very good question. When you look at XtalPi, as you know that a lot of pharmaceutical companies.

Operator: Sorry, we cannot hear the management very clearly.

Speaker #2: So, when we look at the development history—when you look at XtalPi regarding the platform capability of large molecules—we have the large data capacity, and also we have seen the recent years’ development in the AI large molecule space.

Jian Ma: When we look at the development history, when you look at XtalPi, regarding the platform capability of large molecules, we have the large data capacity. We have also seen the recent year development of AI large molecule space. Leveraging the AI capability and big data, we have advanced tremendously. Well, it is not just a methodology, when in reality, it is very important to build this large-scale data, including the industry know-how and real experimental data as a foundation for better model development and iteration. When we look at the course of development for specific targets, we can do larger optimization, substantial optimization from the source. As you mentioned in your question, this is one possibility, and internally, we have accumulated lots of data.

Speaker #2: Leveraging AI capabilities and big data, we have advanced tremendously. Well, and it's not just a, you know, methodology, but in reality, you know, it's very important to build this large-scale data, including industry know-how and real experimental data, as a foundation for better model development and iteration.

Speaker #2: So, when we look at the course of development for specific targets, we can do larger optimization—substantial optimization—from the source, as you mentioned in your question.

Speaker #2: This is one possibility. And internally, we have accumulated lots of data regarding our pipeline development with AI. As people see and discuss in the public market, I think I can encourage everyone to pay more attention to our work related to large molecules.

Jian Ma: Regarding our pipeline development with AI, as people see and discuss in public market, I think people can, I encourage everyone to pay more attention to our work related to large molecule. All right. Thank you, Dr. Ma. Looking forward to the company's next steps of pipeline development. All right. Due to the time constraint, we are going to take one last question. The last question comes from the investor with the phone number ending 2283. Hi, everyone. My name is Yaming from GF Securities. The company divides autonomous scientific discovery into 5 levels based on current progress. When do you expect to reach level 5, general autonomous AIfS? Basically, you are talking about these very high generalization capability and also autonomous decision-making capabilities across various fields. Thank you. Let me address this question.

Speaker #2: All right, thank you, Dr. Ma. Looking forward to the company's next steps in pipeline development. Due to time constraints, we're going to take one last question.

Yu Jun: All right. Thank you, Dr. Ma. Looking forward to the company's next steps of pipeline development.

Operator: All right. Due to the time constraint, we are going to take one last question. The last question comes from the investor with the phone number ending 2283.

Speaker #2: The last question comes from the investor with the phone number ending in 2283. Hi, everyone. My name is Yamin from GF Computer. The company divides autonomous scientific discovery into five levels based on current progress.

[Analyst] (GF Securities): Hi, everyone. My name is Yaming from GF Securities. The company divides autonomous scientific discovery into 5 levels based on current progress. When do you expect to reach level 5, general autonomous AIfS? Basically, you are talking about these very high generalization capability and also autonomous decision-making capabilities across various fields. Thank you.

Speaker #2: Do you know when you expect to reach level five general autonomous AI for science? Basically, you're talking about these very high generalization capabilities and also autonomous decision-making capabilities across various fields.

Speaker #2: Thank you. Let me address this question. First of all, we feel very excited about AI for science as well. We are very optimistic about the future.

Shuhao Wen: Let me address this question.

Shuhao Wen: First of all, we feel very excited about AIfS as well. We are very optimistic about the future. Our expected optimistic timeline is 3 to 6 years of development. The reason for that is because, first of all, if you look at the industry development, it has gathered a lot of attention. Everyone is being excited about the industry, no matter if it is the market, the capital market, or talent development. As we mentioned, a lot of industry players and giants have invested significantly into this field. We also are very excited and optimistic about the future, and we believe that in 3 to 6 years, we can achieve level 5, but we face lots of challenges. First of all, high-quality scientific data still is rare. It is also fraught with many flaws, and high-quality newly generated incremental knowledge is insufficient.

Speaker #2: You know, our expected optimistic timeline is three to six years of development. The reason for that is because, first of all, if you look at the industry development, it's gathering a lot of attention.

Speaker #2: Everyone is excited about the industry—no matter if it's the market, the capital market, or talent development. As we mentioned, a lot of industry players and giants have invested significantly into this field.

Speaker #2: We also are very excited and optimistic about the future, and we believe that in three to six years we can achieve level five. But we face lots of challenges. First of all, high quality scientific data still is rare.

Speaker #2: And it is also fraught with many flaws, and high-quality, newly generated incremental knowledge is insufficient. Also, the platform—by which I mean the industry—also lacks a complete and fixed system for choosing safe platforms.

Shuhao Wen: Industry also lacks a complete and fixed system for choosing safe platforms or stable platforms, and there is lack of industry standards. Secondly, there is still a gap between AI-generated scientific hypothesis and real experiments as well. Based on that, XtalPi also has invested significantly to address these pain points. Everything boils down to the robotics development and AI labs. Secondly, the second effort revolves around large-scale automated AI-driven labs that can generate a great deal of new knowledge, high-quality scientific data as well. This is also called the large model for self-purification or a self-purified large language model. That is why we actually launched the XtalPi Science platform. We look forward to enabling more customers for global R&D so that more researchers around the world can actually use high-quality data for verification and purification.

Speaker #2: Or stable platforms, and there's a lack of industry standards. Secondly, there is still a gap between AI-generated scientific hypotheses and real experiments as well.

Speaker #2: And so based on that XtalPi also has invested significantly to address these pain points. Everything boils down to the robotics development and AI labs secondly we need to you know the second efforts revolves around large scale automated AI driven labs that can generate a great deal of new knowledge high quality scientific data as well.

Speaker #2: And this is also called the large model for self-purification, or the self-purified large language model. That is why we actually launched the XtalPi Science Platform.

Speaker #2: We look forward to enabling more customers for global R&D, so that more researchers around the world can actually use high-quality data for verification and purification.

Speaker #2: So everyone can work together to promote it for better evolution to level five. And we firmly believe that what can be achieved here is huge, with lots of potential.

Shuhao Wen: Everyone can work together to promote it for better evolution to level 5. We firmly believe that what can be achieved here is huge, with lots of potential. Overall, we are very encouraged and excited about this scientific autonomous capability development, and we are witnessing the great birth of an era of great scientific discovery. That is why this field has attracted so many investors, and also governments around the world. I think this can definitely be called a new industrial revolution, which can allow humanity to actually evolve into the next stage of greatness. Thank you. All right. Thank you. Very clear. We are also very optimistic about the future. We also look forward to XtalPi's future achievement in level 5. Thank you so much for everyone's question. Today's meeting has run for 1.5 hours. The discussion has been very thorough.

Speaker #2: Overall, we are very, very encouraged and excited about this scientific autonomous capability development, and we are witnessing the great birth of an era of great scientific discovery.

Speaker #2: And that is why this field has attracted so many investors, as well as governments around the world. I think this can definitely be called a new industrial revolution, which could allow humanity to actually evolve into the next stage.

Speaker #2: Of greatness. Thank you. All right, thank you. Very clear. We are also very, very optimistic about the future, and we also look forward to XtalPi’s future achievement in level five.

[Analyst] (GF Securities): All right. Thank you. Very clear.

Sun Choi Chung: We are also very optimistic about the future. We also look forward to XtalPi's future achievement in level 5. Thank you so much for everyone's question. Today's meeting has run for 1.5 hours. The discussion has been very thorough.

Speaker #2: Thank you so much for everyone’s questions. Today’s meeting has run for one and a half hours. The discussion has been very thorough. I’ll now invite Dr. Xu Hao to give closing remarks.

Shuhao Wen: I will now invite Dr. Shuhao Wen to give closing remarks. Thank you so much for your time today. It is over 1.5 hours. Thank you so much for your attention, and your continuous support of the company. Looking back at H1, we continued to increase investment in core technology R&D. Organic business maintained strong growth. AI for science intelligence solutions achieved a high-speed breakthrough. The XtalPi Science platform and Genius Agent suite were officially launched. Multiple business lines in pharma and advanced materials continue to land important collaborations, further validating our closed-loop technology capability. As you can see, our pipeline assets, compared to a year ago, have improved significantly in terms of the quality and quantity. Looking ahead, we will continue to improve the XtalPi Science platform.

Sun Choi Chung: I will now invite Dr. Shuhao Wen to give closing remarks.

Speaker #2: Thank you so much for your time today. It's been over one and a half hours. Thank you so much for your attention and your continuous support of the company.

Shuhao Wen: Thank you so much for your time today. It is over 1.5 hours. Thank you so much for your attention, and your continuous support of the company. Looking back at H1, we continued to increase investment in core technology R&D. Organic business maintained strong growth. AI for science intelligence solutions achieved a high-speed breakthrough. The XtalPi Science platform and Genius Agent suite were officially launched. Multiple business lines in pharma and advanced materials continue to land important collaborations, further validating our closed-loop technology capability. As you can see, our pipeline assets, compared to a year ago, have improved significantly in terms of the quality and quantity. Looking ahead, we will continue to improve the XtalPi Science platform.

Speaker #2: Now, looking back at the first half, we continued to increase investment in core technology R&D. Our organic business maintained strong growth. AI for Science Intelligence solutions achieved a high-speed breakthrough.

Speaker #2: The XtalPi science platform and Genius agent suite were officially launched. Multiple business lines in pharma and advanced materials continue to land important collaborations, further validating our closed-loop technology capability.

Speaker #2: As you can see, our pipeline assets, compared to a year ago, have improved significantly in terms of both quality and quantity. Looking ahead, we'll continue to improve the XtalPi science platform, with Genius agent as the core orchestration hub.

Shuhao Wen: With Genius Agent as the core orchestration hub, we will further integrate proprietary scientific models, specialist tools, R&D workflows, proprietary data, and physical AI robotic labs. We will keep raising scientific model accuracy, agent decision capability, and cross-scenario reuse efficiency, and drive scale application of AI for science in more fields. We believe that as the value of a scientific intelligence infrastructure is gradually released, XtalPi is well-placed to become an important builder of the global AI for science era, and to create long-term sustainable returns for shareholders. Greater patience is required for the industry before it erupts, but the potential is immense. I would like to thank you again for your patience and support. We are committed to generate long-term sustainable return for our shareholders. Thank you. Thank you so much, Dr. Wen, for the closing remarks. Thank you again to all investors and analysts for joining us today.

Speaker #2: We will further integrate proprietary scientific models, specialist tools, R&D workflows, proprietary data, and physical AI robotic labs. We'll keep raising scientific model accuracy, agent decision capability, and cross-scenario reuse efficiency.

Speaker #2: And drive scaled application of AI for science in more fields. We believe that as the value of scientific intelligence infrastructure is gradually released, XtalPi is well placed to become an important builder in the global AI for science era.

Speaker #2: And to create long-term sustainable returns for shareholders, greater patience is required for the industry before it erupts, but the potential is immense. I would like to thank you again for your patience and support.

Speaker #2: We are committed to generating long-term, sustainable returns for our shareholders. Thank you. Thank you so much, Dr. Wen, for the closing remarks. Thank you again to all investors and analysts for joining us today.

Sun Choi Chung: Thank you so much, Dr. Wen, for the closing remarks. Thank you again to all investors and analysts for joining us today.

Speaker #2: If you have any further questions or would like to continue the discussion please feel free to contact our team this concludes today's call. We wish you a very pleasant day.

Sun Choi Chung: If you have any further questions or would like to continue the discussion, please feel free to contact our team. This concludes today's call. We wish you a very pleasant day. Thank you.

Speaker #2: Thank you.

Operator: Goodbye

Browse all earnings call transcripts

Q2 2026 XtalPi Holdings Ltd Earnings Call

Request TrialDemo

XtalPi Holdings

Earnings