The "legion" of intelligent agents has entered clinical trials. Four Peking University alumni raised $200 million in four months and the valuation increased sixfold.
Text by | Hu Xiangyun
Edited by | Hai Ruojing
"The medical and health industry is experiencing its ChatGPT moment." At the 2026 GTC Conference in March, Huang Renxun mentioned again that the application of AI in the new drug development scenario is bringing huge opportunities.
Recently, foreign media reported that BaiTu Bio and HuaSheng Intelligence have initiated the IPO process. Among Chinese innovative drug industrial chains, "AI-native companies" that have achieved phased results are favored by capital. Among them, the financing pace of Deepwise, an AI clinical R & D service company, has attracted more attention.
From December last year to March this year, the company successively announced three rounds of Series D financing, with a cumulative fundraising of nearly $200 million. It is understood that Deepwise's current valuation has reached billions of dollars, more than six times the valuation in the middle of last year. Moreover, it is advancing a new round of Pre-IPO financing.
According to the company's introduction, the company has seen rapid growth in order volume and performance in the past two years. Since 2025, the sales order amount has been nearly 1 billion yuan. The orders in the first two months of 2026 have exceeded the total for the whole year of 2025. And the company has achieved profitability for many consecutive years. In the recent financing, old shareholders such as Sequoia Capital, CDH Baifu, and Xinding Capital have almost all increased their investments. Each round has "achieved oversubscription and quick delivery".
In the early stage of drug discovery, using AI to optimize drug molecular structures has been widely integrated into the R & D workflow of pharmaceutical companies. To improve the success rate of drug development, the industry hopes that AI can play a greater role in the later clinical trial stage. Today, the technology and capabilities of AI agents have made the industry feel the potential of "silicon-based scientists" to accelerate clinical trials.
Deepwise was founded in 2017. Its founder and CEO, Li Xing, graduated from the School of Pharmaceutical Sciences of Peking University. He has worked in companies such as Johnson & Johnson and Pfizer and has more than 10 years of experience in new drug development in multinational pharmaceutical companies. In the early days, Deepwise entered the market through medical writing scenarios, providing services such as writing clinical trial documents for pharmaceutical companies. Over the past decade, the company has experienced the ups and downs of technological evolution and capital cycles.
Some industry insiders who had early contact with Deepwise mentioned that before the emergence of large language models in 2022, medical documents still needed manual verification after being processed by machines. The unit price was low, and the operating efficiency was not high. But now, the AI + CRO business driven by "AI agents" has a completely different look.
"All along, Li Xing has a concept: 'Everything is text'. The entire process of drug R & D is connected by knowledge, text, and data. In various links such as submitting clinical trial applications and registering for market launch, the carrier of interaction is also text. But what we really focus on is not writing a certain document faster, but how to organize this evidence chain more efficiently and accurately." Chen Ge, the vice president of Deepwise's business, told 36Kr.
Currently, Deepwise has built an "AI agent system" and attempts to autonomously execute and deliver various texts such as clinical trial protocols. What's more imaginable is that the process of delivering clinical tasks is also a process of training and iterating AI, enabling it to have self-evolution and strong insights in a large amount of clinical trial data.
Before the arrival of the AGI wave, Deepwise was dormant for many years. When the technological window opened, the previously accumulated data and understanding of industry pain points allowed it to take a seat at the table. So, why is Deepwise sought after by capital? What are its current business models and core capabilities?
The "Legion" of AI Agents for Clinical Trials
Fundamentally, clinical trial business involves a large amount of data collection, analysis, and management, which requires the cooperation of many experienced professionals and repeated refinement. Complex collaboration, long duration, and high cost are the main pain points in the clinical development of pharmaceutical companies.
In the wave of AI technology, "digitalization of clinical trials" has always been an important application scenario. In the past few years, the clinical trial field has completed the initial informatization and digitalization stages. However, due to the large number of participants and product homogenization under strict regulatory constraints, the market is highly competitive.
However, the demand of pharmaceutical companies persists. If AI models can be used to mine the value of a large amount of clinical trial data, reduce repetitive and cumbersome manual labor, improve output efficiency, and accelerate R & D, pharmaceutical companies are still willing to pay. Market Research Future, a market research institution, said that by 2035, the global AI clinical trial market will exceed $24.37 billion.
Especially after 2022, with the explosion of generative AI represented by ChatGPT, the technological singularity has arrived. Deepwise also began to integrate large models and self-develop an AI-native multi-agent system. The know-how accumulated through delivering medical text projects in the past has been further engineered into system capabilities.
According to data provided by Deepwise, it has served more than 1,000 pharmaceutical companies and delivered more than 40,000 projects.
"This is a rather ingenious idea," a R & D personnel from a multinational pharmaceutical company commented. "The clinical text service team has access to a wealth of data files related to clinical R & D. And when the clinical text service team gets involved, the pharmaceutical company has already completed the more core scientific demonstration and strategy formulation. The working cycle of a single project is not long. This means that they can continuously and quickly access standardized Protocol (clinical trial protocol) and CSR (clinical study report) texts in different R & D fields. It is logical to build an AI system based on this. "
In 2025, in terms of technical architecture, Deepwise built an "AI-native multi-agent system simulating human brain thinking". At that time, this was a concept rarely mentioned by medical companies.
Originally, AI companies mainly relied on increasing the number of parameters and enriching the database to improve the prediction accuracy. However, there is a large amount of information in the pharmaceutical field that is not covered by ordinary databases, which can easily lead to "hallucinations".
Therefore, Deepwise changed its approach. Instead of simply piling up data, it built the system by imitating the human brain's thinking process of "breaking down tasks - continuous feedback" and set up four layers of protection mechanisms for training, reasoning, cross-validation, and post-processing to avoid "hallucinations" from the process design.
Under this framework, a series of complex clinical medical logics included in traditional clinical trials, such as registration and application, trial design, clinical operation, data management, and statistical programming, are broken down into modules in units of "ten thousand", and corresponding agents are developed one by one:
For example, the Protocol Agent responsible for overall clinical protocol design, the Statistical Agent for performing statistical analysis, and the Regulatory Agent for ensuring compliance participate in the production of content such as clinical trial protocols.
Taking Protocol writing as an example, it is the core of the entire clinical trial. In the traditional method, this is a task jointly completed by clinical experts/Principal Investigators (PIs), statisticians, CMC experts, etc., which takes several months to complete. Deepwise's multi-agent system breaks down the Protocol writing task and directly assigns it to Agents playing different expert roles. They understand the trial background and objectives, design the protocol framework, then complete automatic writing, format proofreading, and perform multi-language adaptation as needed.
In this process, not only are the labor and time costs reduced, but the AI agent system may also provide new ideas for human experts based on data accumulation, such as automatically generating materials for regulatory communication meetings.
The know-how provided by AI agents is the technological highlight of Deepwise. The failure of many new drug R & D projects is not because the scientific concept is not advanced enough or the data is lacking, but because there are a large number of gaps between pre-clinical mechanism research and clinical trials, resulting in the drug not being used on the right population, at the right endpoint, or having its potential effectiveness verified through the right protocol.
In this sense, the real value of AI clinical CRO is not simply cost reduction and efficiency improvement, but to be deeply integrated into key clinical links such as patient stratification, enrollment prediction, research center selection, and endpoint optimization, helping the entire clinical development to expose and correct errors earlier. Its essence is not to simply replace manual labor, but to reduce the error rate at key decision-making nodes, improve the efficiency of evidence generation and integration, and thus enhance the closed-loop ability from molecules to patients and from mechanisms to clinical benefits.
Chen Ge introduced that in the early days, Deepwise specifically set up the position of "AI architect" to be responsible for calling different Agents according to specific projects. But now, the system has gradually achieved autonomous calling of Agents and participation in delivery, "and finally, professional personnel review, sign, and approve".
Pay by Results, with a Median Customer Unit Price of Tens of Millions of Yuan
With the implementation of the AI multi-agent system, the business scope that Deepwise can undertake has also expanded significantly as the technology matures. Chen Ge introduced that around 2023, the company's AI system could already deliver typical single-scenario text content such as Protocol and CSR, and the accuracy of the output results was "over 90%". Then it gradually entered the "package business" of clinical trials and undertook CRO orders.
This model of undertaking the whole-process "package service" of clinical trials has directly brought about a transformation in Deepwise's business model, and the company has begun to charge according to the delivery results.
From the most intuitive perspective, this first means an increase in the customer unit price. At the beginning of its establishment, the order volume of Deepwise's single-point services was relatively low, ranging from tens of thousands to hundreds of thousands of yuan. Because it essentially did not deviate from the logic of traditional AI-assisted R & D services, although it had certain advantages in customer acquisition, the revenue ceiling was obvious.
But now, the improvement in service experience and efficiency brought by AI technology has made pharmaceutical company customers willing to "hand over the overall R & D work of the pipeline to us". The median customer unit price has also exceeded tens of millions of yuan. At the same time, since the second half of 2025, the volume of this "package business" has also begun to increase significantly, and "the delivery probability is high, and the efficiency far exceeds that of traditional labor-intensive CRO services".
The medical industry naturally has a strong regulatory background. Therefore, both enterprises and medical institutions tend to be more cautious when facing innovative technologies. The aforementioned multinational pharmaceutical company personnel also mentioned that although AI modular writing is already a relatively practical application and is now used within large pharmaceutical companies, at present, in actual operation, they are more inclined to let it write repetitive content, such as filling in content and corresponding professional expressions in the context. "They are not very confident in AI for descriptive content and will be more cautious when using the first draft of the Protocol directly generated by AI."
So, how did Deepwise quickly gain the trust of its partners and achieve order growth?
"We don't engage in price wars," Chen Ge said. "Instead, we gain the trust of customers by improving the pre-sales experience." For example, when initially contacting pharmaceutical company customers, Deepwise will first use AI agents to provide customers with the materials needed to make clinical protocol decisions, such as pipeline research reports, preliminary protocol discussions, and specific analysis and comparisons. It can even continue to revise the protocol based on feedback. "These protocols are not template-based outputs but are determined according to the actual situation of the project. They have a high degree of personalization and completeness. Our system can usually complete a high-quality iteration in a few hours, without significantly increasing the internal burden, and can also allow customers to quickly establish trust in our efficiency and quality."
This way of delivering value first and then discussing cooperation has, to a certain extent, lowered the decision-making threshold for partners. Chen Ge mentioned that many potential customers originally only planned to hand over pipelines with lower priorities to Deepwise for a trial, but after contact and verification, they "are willing to reach a comprehensive strategic cooperation and even hand over all the relevant needs of the pipelines in the next stage as a package".
Chen Ge revealed that recently, the company has just won a large order of nearly 200 million yuan, responsible for the global multi-center clinical CRO business of multiple pipelines of a pharmaceutical company customer. "We expect that the proportion of revenue recognized in the current year can reach over 70%, which will directly drive this year's performance growth." same time, successful implementation of the package business in the clinical trial stage is just the first step Deepwise has also extended its services to the front and back ends of clinical development "The technical capabilities are the same only the business scenarios have changed Theoretically if AI knows which drugs in the clinical trial stage are more likely to be approved it must also be smart enough in the pre-clinical target discovery stage and in the marketing end it will also know how to guide doctors to use drugs rationally" Chen Ge explained
Recently, Deepwise also said that Edvard Moser, a Norwegian neuroscientist and the winner of the 2014 Nobel Prize in Physiology or Medicine, will join the company as a non-executive director, aiming to "explore the capability boundaries of the company's model and agents in the AGI era from more dimensions".
"Currently, many leading AI companies are exploring cross-border directions outside their core businesses. Because in the AI era, the marginal cost of business services is relatively low, and the winner-takes-all effect is more obvious. Its core cost is the token consumption, rather than the expansion in different regions or more cumbersome management issues like traditional enterprises. In the pharmaceutical field, we believe that the cost and process of new drug listing will gradually shorten in the future, and many traditional practices that are difficult to change will be subverted. This is our judgment on industry competition and future development." Chen Ge said.