SK Biopharmaceuticals通过SK Telecom云服务引入128块Nvidia B200用于AI新药研发

SK Biopharmaceuticals is building its own artificial intelligence infrastructure to accelerate drug development. The company plans to improve efficiency across the entire process, from identifying drug candidates to preclinical and clinical research. It marks the first time a South Korean pharmaceutical or biotech company has secured AI accelerators on a large scale through a subscription-based service.
SK Biopharmaceuticals said on the 2nd that it has secured large-scale in-house AI computing infrastructure by adopting 128 units of Nvidia's B200, an AI accelerator based on the Blackwell architecture. The contract, structured as an annual subscription, is worth about 17 billion won. The deployment was carried out through Haein, the country's first B200-based data center cluster, built by SK Telecom.
SK Biopharmaceuticals (326030) plans to use the computing resources for what it calls the AI transformation of translational research, or TR-AX. Translational research is the process of connecting findings from preclinical studies of drug candidates, such as animal testing, to clinical research in humans.
The new AI infrastructure is expected to shorten SK Biopharmaceuticals' drug development timelines. The company has built a development system in which it identifies drug candidates and conducts preclinical research in South Korea, then carries out clinical trials in the United States. Cenobamate, an epilepsy treatment that won marketing approval and was commercialized in the U.S., is a leading example.
The company will also build a dedicated large language model for use across a range of internal operations, aiming to raise productivity throughout drug development, from candidate identification to research decision-making.
Lee Dong-hoon, president of SK Biopharmaceuticals, said the company will maximize its use of AI across the entire drug development process based on world-class B200 infrastructure to improve the speed and efficiency of research and development.
Meanwhile, competition to secure AI infrastructure has intensified overseas, led by global pharmaceutical companies. Eli Lilly built LillyPod, an AI supercomputer equipped with 1,016 Nvidia Blackwell Ultra graphics processing units, in February this year. Roche added 2,176 Nvidia Blackwell GPUs in March, bringing its total Blackwell GPU infrastructure to more than 3,500 units.