Raised over 100 million yuan in financing merely one month after its establishment, Tsinghua University PhD Gao Bowen forays into the AI pharmaceutical sector | Xinglian Portfolio Updates
The dark horse of AI pharmaceutical R&D has arrived.
AI pharmaceutical enterprise AIPher announced the completion of a 100-million-yuan angel round of financing, which is led by Xianghe Capital and joined by institutional investors such as Xinglian Capital. From company registration to the completion of the first round of financing, it only took more than one month. For a startup that is still in the early stage with a small team, such a financing speed fully demonstrates the recognition of the capital market.
From the perspective of the team, the AIPher team has a solid and strong academic background. Gao Bowen, a doctoral student majoring in Computer Science at the Institute for AI Industry Research, Tsinghua University, serves as the CEO. Lan Yanyan, Vice President and Chief Researcher of the Institute for AI Industry Research, Tsinghua University, serves as the Chief Scientist. Zhang Yaqin, Chair Professor of Intelligent Science at Tsinghua University and Founding Dean of the Institute for AI Industry Research, serves as the Chief Scientific Advisor. In addition, the core members also include doctors and assistant researchers from the School of Life Sciences of Tsinghua University, as well as general managers of well-known biotech firms. Before its establishment, the team had accumulated a number of scientific research achievements around AI pharmaceutical R&D, including DrugCLIP published on *Science*, and PharmAgents for full-process drug R&D.
From the perspective of the track, in the past few years, AI pharmaceutical R&D has gone through the process from the rise of concepts to industrial verification. Leading companies such as XtalPi and Insilico Medicine have successively entered the capital market. AI has begun to evolve from an auxiliary tool in the traditional pharmaceutical R&D process to core R&D links such as target discovery, molecular design, and virtual screening. At the same time, AI for Science is becoming an important direction of scientific research, which has shifted from making existing processes run faster to trying to redefine scientific discovery itself.
The self-developed PharmAgents agent system of the company enables AI Agent to become the "decision-making brain" of drug R&D. It is not only responsible for model invocation and task planning, but also tries to participate in experimental decision-making and result analysis, and re-inputs real experimental feedback into the system to form a closed loop of "model - decision - experiment - data - re-learning". This means that what AIPher intends to build is not a faster drug screening tool, but a brand-new drug discovery engine.
This company defines itself as an AI-native pharmaceutical company. This concept is also incorporated into the company's English name "AIPher": among them, "AI" stands for artificial intelligence, and "ph" not only corresponds to "Pharm" (pharmaceuticals), but also represents "Philosophy". From AIPher's perspective, AI Native does not simply mean applying AI technology to traditional pharmaceutical R&D processes, but rethinking drug R&D from the perspective of AI, and further implementing the reconstruction of organizational modes and workflows.
Tsinghua scientists embark on entrepreneurial journey
Before the company was officially established, this team from Tsinghua University had been continuously exploring in the field of AI pharmaceutical R&D for nearly five years.
The core team comes from institutions including the Institute for AI Industry Research and the Department of Computer Science of Tsinghua University. Lan Yanyan, the Chief Scientist, is Wanguo Data Professor of Tsinghua University, Vice President and Chief Researcher of the Institute for AI Industry Research. She has been deeply engaged in the fields of artificial intelligence and natural language processing for 20 years, has published more than 100 papers in important international journals and conferences, with over 14,000 Google Scholar citations, and her related achievements have been published in the main issues of *Nature* and *Science*.
The Chief Scientific Advisor is Zhang Yaqin. As Chair Professor of Intelligent Science at Tsinghua University and Founding Dean of the Institute for AI Industry Research, Zhang Yaqin also holds the titles of Foreign Academician of the Chinese Academy of Engineering, Academician of the American Academy of Arts and Sciences, Foreign Academician of the Australian Academy of Technology and Engineering, and Academician of the International Eurasian Academy of Sciences, and has long been engaged in research in fields including artificial intelligence.
The academic background of this team also forms the initial technical foundation of AIPher. It is reported that the team began to systematically invest in AI pharmaceutical research in 2021, continuously accumulated technologies around the combination of AI and drug R&D, and gradually formed a series of core technologies including DrugCLIP. Since 2021, more than 30 papers have been published, covering top international AI conferences such as NeurIPS, ICLR, and ICML, as well as top international journals such as *Nature*, *Science* and their sub-journals.
What drives this scientific research team to start a business is their judgment on the intersection of AI and life sciences. Lan Yanyan introduced that the team had started to promote the work related to agents in the second half of 2024. At that time, the team judged that with the further improvement of AI capabilities, agents could not only handle traditional tasks such as language and vision, but also potentially enter the fields of drug discovery and broader scientific discovery to undertake complex research work that was previously mainly completed by scientists.
In 2025, the capabilities of large models such as DeepSeek were seen by more people, which further reinforced this judgment. For the team, the leap in large model capabilities does not simply mean that "AI has become smarter", but makes them see the possibility of AI entering the real scientific world.
Lan Yanyan compares drug R&D to "the pearl on the crown of life sciences". Different from problems in AI fields such as natural language processing, real drug R&D not only requires AI to understand and predict living systems, but also to further complete intervention, find a suitable molecule, and prove that it can produce expected effects in complex living systems. "If we can successfully develop the drug, it means we have a complete understanding and characterization of the entire living world." She believes that this complexity precisely means that there is room for new technological breakthroughs in the traditional R&D paradigm, which also becomes the reason why AI must enter this field.
The technologies accumulated by the team before have also begun to move from papers to real drug R&D tasks. In 2023, the team completed the core technology of DrugCLIP, and successively carried out experimental verification around multiple different targets. At the same time, the team continued to expand the capabilities of AI in different links of drug R&D and different drug modalities, and relevant achievements were continuously published in top international AI conferences and top journals such as *Nature* and its sub-journals. In January 2026, the relevant achievements of DrugCLIP were published in the main issue of *Science*, and the large-scale virtual screening capability for human genome-level targets was further realized. On the basis of the accumulation of these professional models and real projects, the team further developed the full-process agent system PharmAgents for drug R&D, which was first released at the 2025 Hong Kong AI for Science Summit, enabling AI to gradually move from completing single-point tasks to scientific exploration and R&D decision-making. Recently, the team will also release the technical report and evaluation of PharmAgents 2.0 to further demonstrate the capability of AI to participate in real drug R&D.
Gao Bowen introduced that the company has previously carried out experimental verification around multiple targets. Some cases have more overlaps with the structures in the training data, which are relatively easy cases; some have obvious differences with the training set, which are used to test the model's ability to face "out-of-distribution" tasks; there are also more complex cases where only protein sequences existed before, no known co-crystal structures bound to molecules, so it is necessary to predict the structure according to the sequence first, then find potential pockets and carry out molecular screening. In the well-known international drug virtual screening competition CACHE, the organizer provided challenging targets, and each participating team designed molecules and submitted them uniformly. The organizer completed synthesis and activity tests. The molecules screened and submitted by AIPher based on DrugCLIP and other core technologies became the only one among more than 1,300 molecules in that competition that was verified to have activity in three different experimental systems.
Another case closer to the real drug pipeline comes from the ADHD direction. When the team studied the target NET in scientific cooperation, they combined AI capabilities to discover a brand-new binding pocket that had never been reported before and had not been used for drug development. The importance of this discovery is that the new pocket can bring higher target selectivity and allosteric effect, thus forming a differentiated advantage for subsequent drug design, and is expected to reduce the side effects of existing marketed drugs. Based on this discovery, AIPher further advanced the relevant pipeline. The company combines AI capabilities with the experience of human experts to continuously optimize the molecules, and has now advanced to the preclinical candidate compound confirmation stage.
Gao Bowen revealed that at present, AIPher has established more than 30 cooperations with pharmaceutical companies, hospitals, universities, and scientific research institutes, and continuously verified the ability of the model to discover effective molecules on multiple targets. In March 2026, AstraZeneca and Tsinghua University officially announced the establishment of the "Tsinghua University (Institute for AI Industry Research) - AstraZeneca Joint Research Center for AI Drug R&D". The two sides will carry out cooperation around AI-driven molecular research, translational medicine, and clinical development. For this team, this means that the exploration of AI pharmaceutical R&D has further moved from laboratory research to the R&D system of large pharmaceutical companies.
Financing exceeded 100 million yuan just over a month after establishment
After the exploration of AI pharmaceutical R&D in previous years from concept verification, platform construction to drug pipeline implementation, the focus of capital attention is changing: whether AI can truly enter the drug R&D process and finally precipitate into realizable drug assets.
Xianghe Capital, the lead investor of this round of financing, has made multiple bets in the AI for Science field, and has previously laid out enterprises including HuaDeep Pharma, BioMap, and SimpleX. Among them, HuaDeep Pharma is one of the earlier AI pharmaceutical projects that Xianghe Capital bet on. Xianghe Capital entered from the angel round and continued to increase its investment; in 2026, HuaDeep Pharma completed multiple rounds of financing of 787 million US dollars, and reached a strategic cooperation with Sanofi with a maximum scale of 2.56 billion US dollars.
Gao Bowen introduced that on the one hand, AIPher will promote its self-developed drug pipelines in the future, and on the other hand, it will cooperate with leading domestic and foreign pharmaceutical companies. However, the cooperation will not stay in the technical service mode common in traditional AI pharmaceutical companies, but will tend to co-development. "We will not charge a one-time technical service fee." In his view, if only one technical service is completed, it will not help much in the accumulation of the company's technical system and capabilities.
What AIPher hopes to obtain is the rights and interests of the drug assets themselves. Specifically, the company's future revenue can come from R&D milestone payments, pipeline license-out, and equity sharing after the final marketing of the drug. In other words, AI capability itself is not the final commodity that AIPher wants to sell, and what it really wants to precipitate is the drug assets generated by AI capabilities.
This is also AIPher's core judgment on its own business model: AI is not a layer of tool outside the pharmaceutical industry, but should gradually become part of the drug R&D process itself. Gao Bowen believes that AI can first significantly accelerate a single pipeline, especially in the preclinical R&D stage; more importantly, it can change the traditional biotech mode of focusing on one disease field at a time and relying on a few professional teams to advance a single project. With the help of AI, AIPher can promote multiple projects in different disease fields and different drug modalities at the same time.
"A great advantage of AI is that its capabilities can be expanded rapidly. Entering a new disease field or drug modality does not mean that we need to rebuild a new team and capability system from scratch."
AI redefines the organizational mode of drug R&D
As we all know, drug R&D is a long and expensive probability game. In the past few years, the most intuitive value of AI pharmaceutical R&D has been reflected in efficiency improvement. Traditional virtual screening often requires calculation and experiment one by one for a large number of candidate compounds, while AI can quickly complete prediction, screening and sorting in a larger molecular space, compressing part of the calculation work that originally takes months or even longer to several days or even shorter. Taking DrugCLIP previously released by the Tsinghua team as an example, its research results show that AI-driven virtual screening can increase the speed of traditional screening by a million-fold.
But speed may only be the first level of change. What really determines the value of AI pharmaceutical R&D is whether it can improve the probability of "finding the right molecule". This is also the key for the current AI pharmaceutical industry to move from "AI-assisted R&D" to "AI-native R&D". In the past, AI was more used as a tool in the R&D process, such as helping researchers predict protein structures, screen compounds, and analyze experimental data; the further development of AI pharmaceutical R&D hopes to let AI participate in or even redefine the entire decision-making chain from target discovery and molecular design to experimental verification and iterative optimization.
What AIPher hopes to take is exactly the latter path. Gao Bowen introduced that the company's current technical framework includes three core parts. The first layer is the ability to build proprietary models based on public data, further forming "Skills" for different tasks of drug discovery on the basis of large models, to carry out large-scale exploration in the vast chemical and mechanism space. The second layer is to build a decision-making system around the entire pharmaceutical process, so that AI not only completes single-point tasks, but also understands how different R&D links are connected and what to do next. The third layer is the feedback from real experiments.
Experiment is not the end of the AI pharmaceutical R&D process, but the starting point of the next round of model evolution. AIPher will submit the results generated by the model to real experiments for verification, and then feed back the results of successful and failed experiments, molecular properties and the experience of scientists to the system. With the continuous progress of experiments, the model capability and decision-making capability are continuously iterated, and finally a data flywheel is formed.
This idea is obviously different from the traditional way for pharmaceutical companies to accumulate data. After decades of R&D, large pharmaceutical companies certainly have massive amounts of experimental data, but these data are essentially "naturally generated" in the existing R&D process. The difference of AI-native companies is that they can think before the experiment starts: what experiment is most worth doing, and what data is most valuable for the next round of model training.
The team even regards "how to obtain the data required by the model" as part of the design of the AI system. Gao Bowen gave an example that the data of the binding structure between proteins and small molecules is very limited, and it is difficult to support model training by relying only on the existing data. Therefore, the team will try to find amino acid fragments similar to the morphology of small molecules from pure protein data, and construct the surrounding regions into structures similar to "pseudo-small molecule - pseudo-pocket", and simulate the real interaction between small molecules and proteins through synthetic data. The same approach is further extended to cyclic peptide data.
The difference in this idea is closely related to the organizational mode of the team. From Lan Yanyan's perspective, AIPher is neither a pure AI company nor a traditional biotech. The traditional mode may be that the AI team completes the calculation first, then hands over the results to the medicinal chemistry team; after the medicinal chemistry team completes the next step, it is handed over to the biology team for verification. There is an obvious "handover" between different links. AIPher hopes to break this separation. AI scientists, medicinal chemistry experts and biologists jointly discuss problems from the very beginning: what problems are worth solving, what experiments should be designed, what the experimental results mean, and how the next round of models should be adjusted.
Gao Bowen admitted that at this stage, AI has shown good capabilities in discovering Hit/Lead molecules with good affinity and activity, but further advancing from Lead to PCC and even clinical candidate drugs still requires continuous technological improvement.
This precisely means that the real industrial competition of AI pharmaceutical R&D may have just begun.