Article · AI translation
IT Home reported on November 15 that, according to a report today by South Korean media outlet Etnews, South Korean telecom operator LG U+ will enter the AI Token business next year, using its independently developed Token optimization technology to help businesses reduce the cost of AI transformation (AX).
According to the report, the service has three technologies: “Harness Engineering” for designing the paths by which Agents carry out tasks, “intelligent routing” for assigning models based on the type of question, and a “chip combination” that uses heterogeneous semiconductors to optimize inference.
IT Home learned from the original report that LG U+ will reduce unnecessary repeated searches and verification when AI Agents perform tasks and match each task with the most suitable model based on its difficulty. Combined with model compression and inference technologies, the company's independently developed on-device AI model still achieves more than 95% of the performance of a general-purpose cloud model while being reduced to one-third of its original size. By having a small model generate answers quickly and then a large model verify them, Token generation speed increases by more than 2 times.
On the infrastructure side, LG U+ uses a chip combination strategy, pairing Nvidia and AMD GPUs with neural processing units (NPUs) from South Korean company FuriosaAI to handle question analysis (Prefill) and answer generation (Decode) tasks separately. Separating tasks that require large-scale computation from those with high memory-bandwidth requirements reduces infrastructure costs.
LG U+ will both provide services jointly with partners and build its own services as it makes a full-scale entry into the B2B AI Token market starting next year. With LG U+ joining, all three of South Korea's major telecom operators will have entered the Token economy.