On September 23, U.S. artificial intelligence startup Anthropic announced that its Claude model helped discover a new enzyme system, marking the first achievement of its biology research program.

Just days before this discovery emerged, Reuters reported that Anthropic had quietly established a wet laboratory in the San Francisco Bay Area, expanding its business from AI into the life sciences and drug-related research fields.

With powerful computing capabilities, AI technology can screen and design new drug molecules and delivery methods, discover disease targets, and even organize automated experiments, and has long been highly anticipated by drug developers.

In 2026, AI drug discovery has gradually moved from concept to reality. Some companies have announced that AI-designed drugs have entered Phase III clinical trials, AI model giants have begun implementing drug research and development projects, and multinational pharmaceutical companies have become more aggressive in their AI deployment.

In China, on September 18, the Ministry of Industry and Information Technology and nine other departments jointly issued the "15th Five-Year Plan for Pharmaceutical Industry Development," which for the first time included the "AI-driven full-process data automated collection and traceability system for drug discovery" as a high-value scenario in the national five-year plan, emphasizing the empowerment of innovative drug research through artificial intelligence and quantum computing, sparking industry discussion.

"When AI intelligence reaches a certain level, forward-looking players are all thinking about how to use it as a new kind of 'energy' to reshape the industrial chain," said Zheng Shuangjia, assistant professor at Shanghai Jiao Tong University and researcher at Lingang Laboratory, in a recent interview.

AI Drugs Enter Phase III Clinical Trials, Companies Begin to Profit

Chen Kaixian, academician of the Chinese Academy of Sciences, previously mentioned in a public speech that the prospect of AI application in drug research lies in building a "proactive data-driven intelligent drug discovery new paradigm." First, it combines "dry and wet methods," meaning the "brain" of AI (virtual screening, model prediction) combined with the "hands" of high-throughput automated experimental platforms, forming a rapid closed loop of "design-execute-learn." Second, by constructing digital life forms such as "virtual cells" and "virtual patients," simulating drug effects in computers to predict efficacy and toxicity at extremely low cost, thereby "greatly reducing the failure cost of clinical research."

Since the 1980s, scientists have been using computer-aided drug research. After the rise of deep learning, a batch of companies began using AI for target identification and molecule generation, and the concept of AIDD (AI Drug Discovery and Development) emerged. In 2018, AlphaFold achieved experimental accuracy in protein structure prediction, and since then AI-designed molecules began entering clinical trials, igniting capital enthusiasm, with global AI drug discovery financing exceeding $10 billion in 2021. As the first batch of AI-designed drugs entered clinical trials, the industry entered a cooling-off period, with some pipelines failing and IPOs facing cold reception. The market realized that AI models can shorten the time for drug molecule design and screening, but their output still needs to face clinical validation. Although different definitions of "AI drugs" may lead to discrepancies in statistical counts, overall, the number of drugs entering Phase III clinical trials is currently very limited. According to a ranking compiled by Arterial Network on September 14 based on ASCO 2026/JCO research, among 63 companies and 117 AI drug pipelines entering human clinical trials, 60 have completed Phase I, but only 8 have completed Phase II, and 3 have entered Phase III.

With the rapid iteration of AI technology, AI drug discovery seems to be welcoming a new turning point. On September 10, Insilico Medicine's independently developed idiopathic pulmonary fibrosis drug Rentoserib completed the first Phase III patient enrollment and dosing at Peking Union Medical College Hospital and Shanghai Pulmonary Hospital. The drug's target and molecule were entirely designed by AI, with top-line data expected to be read out in 2029. Meanwhile, Deltrix Pharma's MDR-001 for metabolic diseases has advanced to Phase III, with Phase IIb data showing a maximum weight reduction of 10.3% at 24 weeks. Relay Therapeutics' RLY-2608 demonstrated a median PFS of 11.1 months in breast cancer clinical trials. Generate:Biomedicines' AI antibody GB-0895 also has two global Phase III trials underway.

Along with pipeline research progress, AI drug discovery companies are gradually shedding their reliance on capital "transfusions" and beginning to profit. In August, Insilico Medicine released its half-year financial results, achieving revenue of $106.3 million, a year-on-year surge of 287.2%, with a gross margin as high as 90.3%, becoming the first AI drug discovery company listed in Hong Kong to achieve comprehensive half-year profitability. The total value of licensing partnerships it has publicly announced during the year has reached nearly $7 billion, with partners including Eli Lilly, SK Biopharmaceuticals, Takeda Pharmaceutical, and Servier.

Tech Giants Accelerate Deployment

Leading AI model companies with high-level general-purpose models have strengthened their biopharma deployment this year. In April, OpenAI established a strategic partnership with Novo Nordisk, subsequently launching the GPT-Rosalind model for complex mechanism reasoning in biopharma to accelerate disease target analysis and screening. NVIDIA officially launched the BioNeMo Agent Toolkit in June, which has now been integrated by over 50 industry leaders including Eli Lilly, Dassault, and Schrödinger. Earlier in January, NVIDIA announced a 5-year, $1 billion collaboration with Eli Lilly to jointly build an AI innovation laboratory.

In various stages of biopharma research and development, how to more efficiently validate AI-screened and designed drug molecules in the laboratory is a major challenge for AI application. Leading AI drug discovery companies have been actively promoting the construction of automated laboratories in recent years, automatically conducting experiments and feeding results back to AI, forming a "dry-wet closed loop" for drug exploration. This year, leading AI model companies have also begun entering this track.

On September 18, Anthropic officially established a wet laboratory in the San Francisco Bay Area and recruited biochemistry scientists. The project aims to use its large model Claude to direct automated robotic units in conducting biological experiments. Previously, Anthropic had spent approximately $400 million this year to acquire automated experiment platform Coefficient Bio, and released the scientific intelligent agent control standard MHS.

Model companies entering biopharma, "advancing the dry-wet closed loop and letting models 'grow' physical form, could be an industry disruption," Zhang Xincheng, founder and CEO of Xingyuanhui, told The Paper. Leading model companies hold large amounts of capital and can conduct large-scale data collection to feed back into scientific model research and development, ultimately perhaps enabling AI models to reach the astonishing level in biopharma research that they have achieved in coding. "The entry of giants both validates the dry-wet closed-loop AI research model and indeed advances industry integration, but this process is not achieved overnight. In the field of material discovery including biopharma, the exploration space remains vast, and data accumulation, model maturity, and laboratory infrastructure all require time," XtalPi told The Paper. "Taking a longer view, the competitive dimension will become more diversified, shifting from single model capability to a comprehensive contest of data, experimentation, engineering, and commercialization capabilities. The essence of this integration is the inevitable process of the industry moving from proof of concept to value realization."

Alex Zhavoronkov, founder and CEO of Insilico Medicine, told The Paper that model companies need to produce truly useful drugs to prove themselves in the pharmaceutical field. "Since 2025, tech giants like Google, IBM, and Microsoft have spent enormous amounts of money in this track, but have not produced a single drug," he said. "To use AI to empower drug discovery at scale, you first need to actually discover several drugs before you can expand on that basis. We already have such experience, so we are happy to cooperate with them."

BD Deals Land Densely, Multinational Pharma Companies Focus Deployment

In the past, pharmaceutical giants often viewed AI as a research and development auxiliary tool, but this year, AI has been upgraded to a core pipeline strategy for some pharmaceutical companies. Novo Nordisk partnered with OpenAI in April, and on September 16 deepened its binding with Anthropic Claude; Eli Lilly invested $1 billion with NVIDIA to build a joint laboratory, and signed a $2.75 billion BD deal with Insilico Medicine (Editor's note: BD refers to the entire category of transactions between pharmaceutical companies including license-in, license-out, joint research and development, and co-commercialization, collectively known in the industry as "BD transactions"). In addition, Pfizer deepened its cooperation with Isomorphic Labs, expanding its pipeline to 5 new oncology targets; Roche strategically collaborated with Recursion to accelerate its internally built Lunar intelligent computing platform; Ono Pharmaceutical announced in July the organization-wide scaled deployment of its intelligent agent research platform BiomniLab.

Unicorn companies in the AI drug discovery field, while obtaining massive financing, are deepening cooperation with large pharmaceutical companies. Isomorphic Labs, spun off from Google DeepMind, completed a massive $2.1 billion Series B financing in May, and its self-developed next-generation design engine IsoDDE comprehensively empowers cardiovascular and oncology drug collaborations with Pfizer and Novartis.

A senior executive at a leading multinational pharmaceutical company told The Paper that foreign companies' AI deployment has moved from concept pilot to pipeline realization period. AI has been upgraded from a single-point tool to an enterprise-wide computing power and Agent intelligent base, embedded in the entire process of research and development, clinical trials, and medical affairs. "Foreign companies will choose to cooperate with AI companies, but they maintain sovereignty over clinical trials, registration, and pipelines, focusing on building dry-wet experiment closed loops," she said. "Companies no longer look at demonstration effects, but focus on whether AI-produced molecules can advance to Phase II/III clinical trials and obtain clinical evidence. In addition, regulatory compliance is also being front-loaded, with model traceability and explainability becoming essential requirements."

Regulatory Policies Introduced Successively

The international regulatory system for AI medical implementation has also been intensively introduced this year. On January 14, the U.S. FDA and European EMA jointly issued ten principles of "Good AI Practice in Drug Development," establishing unified guidelines of "human-centered, risk-proportionate, full-lifecycle management." In a warning letter in April 2026 regarding pharmaceutical company automation and model compliance, the FDA explicitly wrote that human accountability for regulations and drug efficacy quality must never be transferred to algorithms. China's National Medical Products Administration issued the "Implementation Opinions on 'AI + Drug Regulation'" this year, explicitly proposing to establish a smart regulatory system by 2030 and conduct AI-based risk profiling and non-on-site flying inspections. In long-cycle, high-threshold mid-to-late clinical stages such as clinical trial design, patient recruitment, and real-world research, the entry barriers for AI are being raised.

"AI biopharma is shedding its conceptual halo and entering a new stage where clinical realization and compliance constraints coexist. Opportunities seem great, but it is not a disruptive revolution, rather a gradual reconstruction of the drug research and development paradigm," the pharmaceutical executive mentioned. "AI will change the efficiency of research and development, but may not necessarily replace pharmaceutical professionals' judgment regarding patients, clinical trials, and scientific risks."