AI pharmaceutical development has broken through critical barriers, driving the success rate of molecular design up to nearly 80%, and capital support and experimental verification are the core prerequisites for challenging "Anti-Moore's Law".
"2026 is a completely different year for biological computing. Three years ago when we talked about this topic, it was still very new; this year, it is probably the hottest topic," Liu Wei, co-founder and CEO of BioMap, said at the recently held 3rd International Biological Computing Innovation Summit.
Over the past two years, the integration of AI and biomedicine has gradually moved from proof of concept to the stage of platform construction and industrial collaboration. The number of cooperation projects announced at this summit has also increased: there were about 3 to 5 signed projects in the previous session, and the number has risen to 11 this year, with the cooperation directions covering fields such as traditional Chinese medicine, virtual cell models for cancer, AI breeding, and overseas market expansion.
A reporter from CLS noticed that as the focus of competition in AI drug development has changed, the discussions at this summit are more concentrated on the practical problems encountered in the industrialization process: after computing models improve the efficiency of molecular design and screening, the industry no longer only needs to deal with the question of "whether candidate molecules can be found", but also includes "who will complete the experimental verification" and "who will bear the long-term investment required for verification".
Verification Becomes a New Bottleneck
Moore's Law in the computing field refers to the fact that computing power is getting stronger and the cost is getting lower. However, new drug R&D is taking the opposite path: The R&D cost of a single new drug roughly doubles every 9 years. Since it runs in the opposite direction of Moore's Law, it is called "Eroom's Law".
Now, this curve is expected to be reversed. Paul Chan Mo-po, Financial Secretary of the Government of the Hong Kong Special Administrative Region, said at the summit that the three forces of progress in computing technology, China's pharmaceutical innovation and long-term capital are bringing opportunities to reduce the cost of new drug R&D. In terms of computing technology, artificial intelligence is assisting researchers to build models of life operation, and has begun to design molecules, antibodies and enzymes, reducing the R&D mode that relies solely on repeated laboratory trials; in terms of pharmaceutical innovation, China has become an indispensable part of the global R&D network. In 2025, the total value of out-licensing transactions of China's innovative drugs approached 140 billion US dollars, about 10 times that of 2021; in terms of capital, despite fluctuations in global biopharmaceutical venture capital in recent years, AI drug R&D platforms still attract a large amount of funds, and this year there are even enterprises that have completed a single round of financing of more than 2 billion US dollars.
But Paul Chan Mo-po also emphasized that artificial intelligence cannot eliminate clinical risks, The R&D bottleneck is shifting from "discovery" to "verification", and future progress depends on data quality, trial speed and regulatory orientation.
Data from front-line enterprises shows that the "design" stage has indeed been successfully completed. Feng Yulan, Senior Vice President of Harbour BioMed, introduced, The company has newly generated 107 completely new molecular sequences through the AI HCAb model de novo antibody platform, and the success rate of binding to targets has increased from about 9% in the past to 78.5%. "This turns the original mode of making choices in a limited space into creating things that did not exist before."
Qu Jiarun, Director of Orbimed, also shares the same view: Five years ago, when doing antibody design, only a few correctly designed sequences could be found from a sequence library of millions of levels. Now, by testing no more than 50 sequences, hit molecules with single-digit nanomolar affinity can be found. "The era of AI has arrived."
As design speeds up, the pressure of verification increases accordingly. Liu Jiang, Vice President and General Manager of Greater China of Revvity, said, The company's high-throughput platform has increased the screening speed by 50 to 100 times, and customers' demand for verification capabilities has also increased by about 70 times. The reason is that in the past, animal experiments helped to screen out a large number of off-target molecules. After switching to computational screening, high-affinity candidate molecules appear in batches, and there are even more that need to be verified.
Since verification has become a bottleneck, completing wet experiments has become a common choice for platform companies. Liu Wei revealed that BioMap is currently selecting a site in Hong Kong for an AI-driven high-throughput wet laboratory, building a joint innovation project of dry-wet closed loop with Revvity, and planning to jointly build an AI biological computing center with Temasek Life Sciences Laboratory in Singapore.
Harbour BioMed is trying another approach. Feng Yulan divides the output of AI drug R&D cooperation into three layers: assets, data and model capabilities. In a single project cooperation, pharmaceutical companies can often only take away assets, cannot take away models and data, and it is difficult to precipitate capabilities.
Harbour BioMed's approach is to jointly establish MegaStream TechBio with BioMap, putting all three elements into the same company. Lei Ming, CEO of MegaStream, proposed that AI should run through the whole R&D process, "not just lab in loop, but everything in loop", so that feedback from clinical and even BD links can be fed back to the model. He calculated that the training parameters of BioMap's base model exceed 260 billion, and a small product company needs at least tens of millions of US dollars to independently replicate it. According to the plan, the company's dry-wet closed-loop laboratory is scheduled to start construction in December, and the first pipeline is planned to submit IND application in 2027.
Liu Wei introduced that MegaStream is not an isolated case. BioMap mainly acts as a technology enabler in the process of incubating projects, which are generally led by large pharmaceutical companies or pharmaceutical investment funds, and more project incubations are currently being promoted.
Long-term Capital Support is the Key
Verification requires long-term investment, and the capital problem follows.
Li Jiaqing, President / Member of the Executive Committee of Legend Capital, reminded that different from the TMT industry, the capital sources of biotech start-ups are not limited to equity financing, and the down payment brought by BD transactions is also an important source of capital. For example, one of Legend Capital's invested enterprises obtained more than 1 billion US dollars in down payment through BD transactions within two years. He judged that in the next three to five years, BD transactions targeting multinational pharmaceutical companies will still be one of the important ways for biotech enterprises to realize pipeline value and obtain capital support; but in the long run, the sustainable development of the industry still needs to rely on the support of the local capital market and local investors.
In terms of capital exit, IPO is still the most mature exit channel at present. Legend Capital has had more than 10 projects completed IPO on the Hong Kong Stock Exchange this year. Li Jiaqing believes that the Hong Kong capital market is becoming one of the important listing and financing platforms. However, Lin Yaojie, President of Capital Strategy of BioMap, pointed out that exit options other than IPO are still very few, and channels such as mergers and acquisitions and continuation funds are relatively limited.
Wang Ting, Founder and Chief Investment Officer of True Pulse Investment, further said that on NASDAQ, the cash brought by M&A of large pharmaceutical companies will flow back to the market to form liquidity; while many Hong Kong-listed biotech companies are repurchasing shares at present, the market value of many companies is close to the net cash level, "which is very rare in the global capital market". Without more flexible M&A, listing and delisting mechanisms, the market may fall into a value trap. The good news is that international investors from the Middle East, Europe and other regions have begun to regard China as a global R&D engine, and their interest in Chinese biotech has risen significantly.
Apart from the exit channels, the shortage of early-stage capital is also obvious. Rong Jing, Managing Director of BlueRun Ventures, believes that the problem does not lie in the total amount of capital, but in the type of capital: biotech investment requires long-term capital, profound scientific judgment and higher risk tolerance. Seed-stage projects lack experimental data, and after entering the clinical stage, the capital demand rises sharply while the risk does not decrease, so many companies with technical potential find it difficult to cross the next milestone.
The structural obstacles make the problem more complicated. Qu Jiarun said that the red-chip structure that Orbimed used to use for incubating enterprises in the past is no longer feasible for listing in Hong Kong since last year, and a new structure needs to be designed.
In response to these problems, Lin Yaojie believes that different stages require different types of capital: early-stage VCs from seed round to Pre-A/A round are more willing to take scientific risks, then industrial parties provide resources and platform verification in Series B and Series C rounds, growth-stage PE is also willing to support companies to promote large-scale development, and later capital with government background provides infrastructure support and policy coordination. The three types of capital are indispensable. Rong Jing suggested that the government act as an anchor investor to contribute to a professionally managed biotech VC fund, while keeping investment decisions market-oriented, using long-term capital to attract better entrepreneurs, better companies to generate higher investment returns, and higher returns to attract more long-term capital, forming a positive cycle.
As Paul Chan Mo-po said at the summit, "Eroom's Law describes the past and does not determine the future." For AI drug development, whether it can convert design efficiency into repeatable experimental verification, and then convert verification results into clinical value and capital returns, is the real exam question for the next stage.
This article is from the WeChat Official Account "ChiNext Observation", author: Wang Junxian, published with authorization from 36Kr.