Tencent Cloud launched an integrated multimodal data processing, model training, and inference solution for physical AI applications on 30 October.
The company announced the release of TCLake XPhysis, a platform designed to cover the entire physical AI engineering chain. Cloud stated that these tasks were previously spread across multiple toolsets, which caused duplicated data preparation and created assets that were difficult to reuse.
The new solution integrates DataBuddy workflows and the AI DLC data lake compute service to host data management, development orchestration, and model jobs on a single platform. Xpark and Meson components handle the labeling, sample preparation, quality checks, and cleaning of multimodal data.
Distributed training and inference are provided by TCRay, while TCLake manages AI and data assets in one location. Cloud claimed that this approach reduced data preparation time from 15 days to just two days. The company also reported that the solution increased GPU utilization by nearly 200 percent and lowered the average daily cost per GPU by 35 percent.
Zhuoyu Technology is among the users named by the company. BioMap utilizes the TCRay and Cloud's EMR DIaaS platform for training and processing multi-omics data for its virtual cell work. Huang Shifei said that the way companies handle computing loads and data is changing as agents become new users of data platforms.
Shifei added that platforms must handle multimodal processing, training, and inference efficiently while running tasks under control and understanding the business accurately. The education company Jiuxuewang moved from a pilot phase to production in three weeks using the technology. A private deployment option is now available for customers in the government and finance sectors.
