[Edaily Reporter Kim Seung-kwon ] Isomorphic Labs, a spin-off from Google DeepMind, raised approximately 2.9 trillion won in a single funding round last May. During the same period, the Series B funding round for a leading South Korean artificial intelligence (AI) drug discovery company amounted to around 16 billion won. A simple comparison shows a difference of about 180 times. This figure encapsulates the current state of South Korea’s AI drug discovery industry. This suggests that the gap lies not in technology, but in “stamina.”

Amid this situation, the global competition in AI-driven new drug development is shifting from a battle over algorithm performance to a contest of “clinical execution capabilities.” Some observers predict that, as a result, domestic companies may ultimately fall behind in the capital race. PharmDaily analyzed the survival strategies of domestic companies as Big Tech begins to actively enter the clinical phase of AI-driven new drug development.

A 2.9 Trillion Battle of Endurance… Clinical Trials and Data Will Determine AI’s Success

Isomorphic Labs’ strategy goes beyond simply providing an AI platform to external parties; it is a vertically integrated model that directly connects the process from candidate compound discovery to clinical trials. The company is conducting joint research on small-molecule new drugs with Eli Lilly and Novartis, and in January of this year, it began a collaboration with Johnson & Johnson (J&J) covering small molecules, antibodies, peptides, and molecular adhesives. The company’s plan is to create a structure where the experimental and development results of AI-designed molecules can be fed back into the model.

The typical path for domestic AI-driven drug discovery companies is different. They use their platforms to identify candidate compounds, demonstrate potential in preclinical or Phase 1 clinical trials, and then license the technology to global pharmaceutical companies to share the risks and costs of subsequent development. From a capital efficiency perspective, this is a rational choice. However, if the company fails to sufficiently accumulate efficacy, toxicity, and patient response data from late-stage clinical trials following a technology transfer—and feed this data back into the model for training—the long-term competitiveness of the AI platform may be limited.

Another gap that the domestic industry is focusing on is data. In a recent interview, Standard Hee, Director of the AI New Drug Research Institute, cited a shortage of skilled personnel, a lack of data, and quality issues as the key challenges identified by domestic pharmaceutical companies and AI-driven new drug development firms. New drug development data is a core asset for each pharmaceutical company, and the more sensitive the data—such as clinical, toxicological, or drug response data—the more restricted its public disclosure becomes.

In contrast, global big pharma companies are securing AI engineers and data scientists on a large scale. Director Pyo assessed that the use of AI is becoming commonplace in big pharma pipelines and that AI is establishing itself as a research infrastructure linking hypothesis formulation, experimental design, and result analysis—going beyond merely shortening specific experimental stages.

The solution lies not in simply consolidating data in one place, but in “data partnerships” where hospitals, pharmaceutical companies, and AI firms collaborate toward clear research objectives. The government is also supporting the integration of preclinical and clinical data through the “K-AI New Drug Development Preclinical and Clinical Model Development Project,” a 37.1 billion won initiative, and is supporting the development of foundation models and AI application technologies based on this data.

The private sector has also begun building closed loops. LGCHEM,LTD has signed a joint research and license option agreement with the UK-based LabGenius Therapeutics to discover multi-antibody anticancer candidates. The process involves AI designing antibody candidates, which are then validated through robot-based experiments, with the results subsequently fed back into the machine learning model. LGCHEM,LTD has set a goal of cutting the time typically required for antibody candidate discovery—which usually takes more than five years—in half and accelerating the transition to the preclinical stage.

Kim Woo-yeon, CEO of Hits and a professor in the Department of Chemistry at KAIST, emphasized, “There is no chance of success if we simply replicate Google DeepMind’s AlphaFold3,” adding, “Since replicas of the model have already emerged in the U.S. and China, we must differentiate ourselves through ideas and technological capabilities.” He continued, “Bio-AI can be applied to various bio-research fields and industries, including new drug development, so securing proprietary technology holds significant economic value.”

Paros Clinical and K-Fold Gain Momentum… Narrowing the Gap Through “Different

Approaches”
Achievements in domestic AI-driven new drug development are emerging in both clinical trials and platform development. One of the most advanced clinical cases is #Pharos iBio Co., Ltd. The company has obtained the final Clinical Study Report (CSR) for the global Phase 1 clinical trial of “PHI-101 (rasmotinib),” a candidate drug for acute myeloid leukemia (AML) identified through its AI-based new drug development platform, “Chemiverse.”

PHI-101 underwent a Phase 1 clinical trial in Korea and Australia targeting patients with relapsed or refractory AML. According to the clinical results released by the company, the rate of comprehensive complete remission (CR, CRi, MLFS) among evaluable patients was 50%, and the objective response rate (ORR) was 67%. While these are early-stage clinical data and require validation through large-scale follow-up trials, it is significant that a candidate compound derived from a domestic AI platform has yielded clinical results in human subjects.

Another candidate compound, “PHI-501,” has also entered the clinical phase. PHI-501 is a pan-RAF/DDR1 dual inhibitor targeting advanced solid tumors with BRAF, KRAS, and NRAS mutations. In January of this year, Pharos iBio Co., Ltd. began enrolling the first patients and administering the drug in a Phase 1 clinical trial in South Korea at Severance Hospital, Samsung Medical Center, and Chilgok Kyungpook National University Hospital. This trial consists of dose-escalation and dose-optimization phases and will evaluate safety, tolerability, pharmacokinetics, and early anticancer activity. The target completion date is November 2028.

Galux is making its mark more in AI-based protein therapeutic design and collaborations with big pharma than in clinical trials. Galux promotes a “de novo” approach to designing new proteins from scratch using its “GaluxDesign” platform, which combines physics-based computing with AI. This year, the company signed a joint research agreement with Boehringer Ingelheim to utilize AI-based protein design technology and is also conducting a new drug development project with AstraZeneca.

However, Galux’s proprietary pipeline is still in the preclinical stage. The company previously presented preclinical results for an AI-designed bispecific antibody immunotherapy at the American Association for Cancer Research (AACR) 2026 conference.

KAIST and HITS’ K-Fold seeks to differentiate itself from global models through “speed and the prediction of structural changes.” K-Fold predicts not only the three-dimensional structure of proteins but also the structures of protein-protein, protein-drug, and DNA-RNA complexes. The research team explained that, unlike existing models that rely on multiple sequence alignment (MSA) to identify similar protein sequences on a large scale before structural calculations, K-Fold replaces this with a pre-training approach, increasing the speed of structure prediction by up to 25 times.

Hitz cited the “shift from discovery to design” as the most significant change AI will bring to new drug development. He noted that research is expanding from the traditional approach—which involved identifying promising candidates among known compounds and optimizing them through iterative experimentation—toward designing new candidates from scratch that meet specific criteria. CEO Kim explained, “While existing models focused on accurately predicting the final structure, we go beyond simple structure prediction to predict structural changes.”