AI-Driven Biotech Renaissance: From Spring to Warring States Era

Li Yun | Author

Wang Chen | Editor

5Month13On [date], METiS TechBio was listed on the Hong Kong Stock Exchange, becoming“AIThe First Drug Delivery Stock”, obtained18Cornerstone Investment Subscription, Surges on First Day126%。

Metis TechBio has always beenClassified under“ChinaAIChina's Four Leading Biopharma Companies”in the tier. Among them, XtalPi2024first to be launched in [year], and is also“18C”its first listed company; Insilico Medicine went public the following year, with the public offering portion recording over...1400times oversubscribed; DeepIntel has secured billions of dollars in financing to date and is currently in thePre-IPORound.

Most of these companies were founded in2015Around the year. At that time, China was just entering an investment boom in innovative drugs. The STAR Market, Hong Kong stocks18AFollowing successive deregulations, substantial US dollar funds and industrial capital from the primary market flooded into biopharmaceuticals; meanwhile,AlphaGoSubsequently,AIhas been endowed with vast room for imagination. And“AI+Biotech”Naturally becomes the crossover track most likely to achieve high valuations.

Therefore, in this batchAICertain commonalities can be observed among pharmaceutical companies: strong narratives, high valuations, delayed realization timelines, and currently uncertain commercial outcomes.

An investor also mentioned that back then, they gave...AIChallenges in Pharmaceutical Valuation:In2020Around the year, manyAIPharmaceutical companies are still unable to even provide sufficient and reliable data.and the aforementionedAIThe "Four Little Dragons" of the pharmaceutical industry have demonstrated relatively strong performance in this regard, enabling them to secure substantial investment.“However, due to a lack of data, the rapid rise and fall of this trend has been exceptionally volatile, with a significant number of companies folding in the process.”

Nowadays, of the previous eraAIThe pharmaceutical narrative is effectively drawing to a close: the surviving companies are already approaching the end of the primary financing cycle.Meanwhile, Frost & Sullivan data shows: globalAIEmpowering pharmaceutical R&D expenditure from2020year's54billion U.S. dollars increased to2024year's137hundred million U.S. dollars, with a compound annual growth rate of26.1%——This means that there will be moreAIAs pharmaceutical companies emerge, the industry shakeout will inevitably become even more ruthless.

According toCB Insightsdata, to2023In [year], globally, there have already been200Multiple startupsAIThe pharmaceutical company is currently in a fundraising phase, and when it2024Over the years, this figure has more than doubled.

Amidst such a flourishing landscape, how should investors make their choices?AIWhat overall changes have occurred in the business models and structures of pharmaceutical companies? What lessons can investors draw from the previous investment bubble and the subsequent recovery?

-01-

Resurgence of Hype or Genuine Prosperity?

If you have gained hands-on experience in this field for over5practitioners with years of experience, it would be impossible not to remember that in2022-23around the yearAIThe bubble crisis in the pharmaceutical industry. Prior to this,AIThe pharmaceutical financing wheel relies almost entirely onAIphafold、Chatgpt、Deepseekdriven by the explosive popularity of large models such as,However, whether for investors or entrepreneurs, this cross-disciplinary field is moreAITechnological development remains far from mature.

2022In that year, a capital winter set in, subsequently sweeping away many enterprises operating under this immature model. Investors recalled that among the startups they engaged with at the time, over 70% were on the verge of cash flow depletion.

At that time, even those four companies that are now industry leaders faced widespread skepticism. XtalPi, at2021also experienced during those years50international institutions' financing bidding, with the quota reaching8hundred million US dollars, post-money valuation reaches19.68hundred million US dollars. However, as2022Due to the slowdown in this year's biopharmaceutical market environment, the company was compelled to shelve its US listing plans.IPOplan. Insilico Medicine's four IPO filing attempts and persistent efforts despite repeated setbacks also reflect the industry's challenges at the time.

This is certainly influenced by the broader macro-funding environment. But setting that aside, the lack of systematic summarization of empirical rules behind the technology, the absence of any final products reaching the market, and the unclear core development trajectory and commercialization pathways are all hindrances.AIReasons why pharmaceutical companies secure financing.

The aforementioned investor noted that, among the earliest batchAIAmid the pharmaceutical investment boom, the roadshow presentations of many companies still focus on“`Re-feed the ground-truth data back into...`AI, thereby testing the predictive accuracy of the model”. Even so, before the bubble burst, many investors still believed in its potential, even though the company's model had yet to demonstrate any forward-looking capabilities.

Even if a group of companies ultimately survive and successfully close a funding round, the aforementioned investors acknowledge that there is still an element of luck involved.“When retrospectively evaluated against current standards, significant misjudgments are not uncommon.”

And since last year,AIBearing their scars, pharmaceutical survivors have once again returned to the mainstream spotlight. Unlike the previous cycle, this time they have securedMNCendorsement.

In the history of biopharmaceutical development,MNCIt often serves as a crucial bridge for the commercialization and widespread adoption of a technology.

2025Year,Isomorphic LabsObtain6hundreds of millions of US dollars in financing, and established collaborations with Eli Lilly and Novartis with a total potential value of up to30billion U.S. dollar cooperation agreement. In China: XtalPi Technology is2025Year andDoveTreeContract Total Price59.9hundred-million-dollar orders; Insilico Medicine has successively entered into pipeline licensing agreements with multiple pharmaceutical companies, and this year also secured an agreement with Eli Lilly with a total value of27.5hundred-million-dollar deals; Metis TechBio also follows a platform licensing model, with the contract value for a single target reaching up to1.09hundred million U.S. dollars.

It could be said that the previous batchAIThe business models forged by leading pharmaceutical companies have set the tone for the investment trends of a new wave of companies.

Another industry investor noted that current target selection will focus more on“Implementable Platform Capabilities”。Rather than focusing on a few isolated instances of successful data validation, investors are more concerned with whether the company possesses the capability to continuously advance multiple projects and rapidly establish partnerships with domestic and international enterprises to deliver results in the short term.

-02-

`From Large Models to Agents`

AIThe development of the pharmaceutical industry relies on a complete and complex industrial chain: upstream relies onGPU, cloud computing, biological databases, and automated instruments; the midstream isAImodels, molecular dynamics, and robotic experimental platforms; downstream connects to pharmaceutical companies,CRO、Hospitals and clinical resources. Among them, the technological capabilities of the midstream sector play the most critical role.

The aforementioned industrial investor mentioned that, since the lastAIIn this current cycle of the pharmaceutical investment boom, the focus of midstream investments has shifted from large models toAgentMetastatic tendency.

so-calledAgent, essentially: capable of autonomously invoking tools, executing tasks, and iterating based on feedbackAIsystem. Unlike previous large models: the former can only process input data and perform reasoning functions; whereas the latter is a closed-loop system capable of multi-step planning, and is therefore also referred to as possessing composite functionalities.“AI Agent”。

At the recently held Annual Meeting of the Chinese Antibody Society, a speaker also mentioned: currentlyAIExcels at literature reviews and well-defined coding tasks, but for scientific questions requiring real-world interaction, human oversight remains essential,AIAs an auxiliary tool. In this regard,AgentIt can minimize human intervention and manual operations to the greatest extent, thereby further improving efficiency.

In recent years, some newly emergingAIPharmaceutical companies will alsoAgentas a key competitive advantage. For example, Wangshi Zhihui focuses on a full-stack AI agent for early-stage small molecule discovery, establishing a closed-loop system encompassing target identification, molecular design, and wet-lab validation; Moshu Digital, meanwhile, possesses nearly 60 For Biomedical Use Only Agent，covering the entire value chain of R&D, project initiation, clinical development, and commercialization.

The aforementioned investors also mentioned that, even in the current context of a second wave of investment enthusiasm, large models on the market that only offer single-function capabilities...AICompetition among pharmaceutical companies remains fierce, making it difficult for them to survive.“It can be said that, if the company does not haveAIintegration plan with wet labs, then it has already been left far behind by the times.In terms of investment, it will be directlypassdrop.”

The fundamental reason for such a change lies in people's [...] over the past few years regarding...AIPharmaceutical observations reveal that a model's generative capability does not necessarily align with clinical success rates.“While the capabilities of large models continue to improve, diminishing marginal returns are already evident. From an investor’s perspective, the potential for further increasing clinical trial success rates in the future lies in enabling more timely data feedback and automating R&D processes.”

3In mid-[Month], Roche announced the expansion of its global artificial intelligence infrastructure, completing in the United States and Europe...2176a high-performanceGPUdeployment; in the same month, Eli Lilly also officially inaugurated its pharmaceuticalAIPlantLillyPod, and announced with NVIDIA that it will invest over within five years10established with hundreds of millions of US dollarsAIJoint Innovation Laboratory.

Two companiesMNCNVIDIA PartnershipBioNeMoThe platform employs a laboratory closed-loop model. Upon large-scale implementation, it will evolve into an end-to-end system, establishing a verifiable loop with wet-lab experiments.

This also means that this round ofAIThe boom will rapidly shift from an investment frenzy to an arms race backed by real capital. Startups must be willing to invest in systematic infrastructure development, so as not to fall out of step withMNCout of sync with the overall pace.

-03-

Challenge the Mechanism

This roundAIThe current pharmaceutical investment boom differs from the previous cycle in another key aspect: capital's focus is shifting from generating molecules to understanding biology.

Over the past few years, large models, diffusion models, and generative chemistry have developed rapidly, enablingAIFor the first time, drug molecules can be designed *de novo* at scale. Many companies are no longer relying on traditional screening libraries; instead, they directly utilize models to generate candidate compounds for rapid optimization through high-throughput screening.

This means that a critical bottleneck is disappearing: high-quality molecular design itself is becoming increasingly less scarce.

In the past, one of the greatest challenges in drug R&D was identifying suitable molecules; but now,AIIt is now possible to generate, screen, and optimize tens of thousands of candidate structures in a single run. Consequently, the industry’s underlying paradigm is beginning to shift. As a speaker at the aforementioned Chinese Antibody Society Annual Meeting summarized: future drug development will be more akin to selecting components from a catalog and assembling them, rather than repeatedly re-engineering from scratch each time.

However, for this very reason, new bottlenecks have emerged.

The industry is gradually realizing: the true challenge lies in understanding exactly what happens to a molecule within a complex biological system. As the aforementioned investor noted:“AIAlthough models excel at predicting molecular properties, they often overlook dynamic factors such as biological activity and tissue specificity. The more complex the target, the more pronounced this issue becomes.”

He also mentioned that, given the current capabilities of companies in China, few convincing solutions to this issue have been observed so far.However, this has not deterred some enterprises from taking the lead in pioneering exploration, while a number of investors have also begun to confidently place their bets.

Amid this wave of investment fervor, industry awareness has undergone a significant shift: a growing number of professionals recognize that what truly determines a drug's success or failure is the depth of understanding of biological systems.

Therefore, an increasing number of studies have begun to emphasize“Physical RestraintAI”. The core idea of this direction is: to integrate traditional physical models, biodynamic models, and deep learning. Because purelyAIAlthough the model demonstrates strong fitting capability, it suffers from poor generalization; it readily performs well on training data yet fails to adapt to novel biological environments. Conversely, while physical models have limited accuracy, they possess stronger interpretability and extrapolation capabilities.

Similar logic is also observed inPK/PDField.IBM ResearchIn2024proposed in a study, usingPhysics-Informed Neural NetworksBy integrating pharmacokinetic differential equations, the model not only fits the data but must also adhere to actual biological kinetic principles.

Guided by this new paradigm, several representative enterprises have begun to emerge in the market: for example, those with molecular dynamics as their core technologySchrödinger, with systems biology as its core technologyRecursionetc., in recent years have all obtained andMNCOpportunities for in-depth collaboration.

In China, however, such enterprises have a relatively short development history and are only just beginning to emerge in the investment market. For instance, Huashen Zhiyao’s incorporation of physical constraints is reflected in,AIWhen generating molecular structures or predicting protein folding, constraining the model to adhere to physicochemical principles such as mechanically stable conformations and electrostatic interactions has improved the success rate of protein design.

Of course, all substantive technological advancements begin with a concept. From2021to2026year, youngAIThe pharmaceutical industry has already undergone two investment cycles. Following the surface-level fluctuations, both the industry ecosystem and market sentiment have accumulated more substantive experience. Concepts are not the original sin of investment bubbles; the key lies in the industry's need to identify more robust anchor points beneath these narratives.

....