Video AI startup TwelveLabs said on Tuesday it has introduced Nvidia's next-generation AI accelerator, the Blackwell Ultra B300, through Amazon Web Services (AWS). It has been operating it in a live research environment since early February. It is seen as a leading adoption case among domestic AI companies.
The B300 is a graphics processing unit (GPU) equipped with HBM3e memory offering 288GB of capacity and 8TB per second of bandwidth. It provides up to 50 percent higher compute performance than the previous generation H200. It is optimised for large-scale model training and high-difficulty inference tasks, and is seen as suitable for developing hyperscale video foundation models.
TwelveLabs plans to combine the computing resources with AWS' distributed training service, Amazon SageMaker HyperPod, to expand research into next-generation video foundation models that deeply understand spatiotemporal context in videos.
TwelveLabs CEO Jaeseong Lee (이재성) said, "We will continue to provide unwavering investment and support for the talent that will set a new standard for video AI on the global stage."
Seul-gi Son
sageson@d-today.co.kr