Lablup is heading into overseas markets by showing integrated AI infrastructure operations technology together with global semiconductor companies. Its strategy is to raise the operating efficiency of AI infrastructure with software that manages different AI semiconductors, such as graphics processing units (GPUs) and neural processing units (NPUs), in a single environment.
Lablup said it will take part in the 'Supercomputing 2026 (SC26)' exhibition, which opens on November 17 (local time) in Chicago in the United States, to present its next-generation AI infrastructure operations technology.
SC26 is a global event covering advanced computing technology such as high-performance computing (HPC), AI, networking and storage. This year's exhibition runs from November 17 to 19 at McCormick Place in Chicago. Lablup is taking part for a fourth straight year and operates booth 4630.
Booth talks and partners
At the event, it will hold 'Booth Talks,' a program of technology presentations in which major companies and institutions from Korea and abroad take part. Six semiconductor companies are participating: the global chipmakers AMD, Intel and Nvidia, and the Korean NPU companies Rebellions, FuriosaAI and HyperAccel.
Including the infrastructure partners AMAX and VAST Data and the University of Southern California (USC), a total of nine companies and institutions are scheduled to share technology and real use cases at the Lablup booth.
Intel will present ways to build agentic AI infrastructure in an open heterogeneous computing environment. Rebellions will present its strategy for operating large-scale AI services based on NPUs, and FuriosaAI will introduce cases of large language model (LLM) operation and power efficiency that combine its own AI accelerator 'RNGD' with Lablup's AI infrastructure management platform 'Backend.AI.'
HyperAccel will share AI infrastructure technology centered on token efficiency. AMD and Nvidia are also scheduled to take part in the technology presentations.
Customer cases
Real customer use cases will also be presented. USC will present a case of using Backend.AI to provide AI computing resources to university researchers. The Korea Institute of Science and Technology (KIST) will introduce a case of research on the Cas9 gene editing mechanism using the Backend.AI cloud.
AMAX, an infrastructure partner, will share ways to adopt AI in on-premises environments, and VAST Data will share ways to optimize the KV cache together with Lablup.
Operations strategy and new products
Lablup will also present an AI infrastructure operations strategy under the theme 'If you have built a token factory, who will run it?' A token factory is a concept that views an AI data center as a factory that produces tokens, the inference output of AI models.
In particular, it will present ways to run LLMs in the closed-network environments of the financial sector. The approach uses Backend.AI to manage different AI accelerators and computing resources in one place, and 'Continuum Enterprise' to manage everything from AI model request handling to token usage, operating policies, costs and billing.
It will also show 'AI:GO,' a desktop application that runs AI models on a personal PC without a cloud server, and full-stack observability technology that checks the state of the whole AI infrastructure.
Lablup will also unveil a 'Slurm frontend' that runs jobs in the Backend.AI container environment while keeping the commands of Slurm, an existing HPC job management tool. It also plans to introduce 'Confidential AI' technology, which uses hardware-based trusted execution environments (TEEs) to protect sensitive data and AI models.
Shin Jung-gyu, chief executive of Lablup, said, "As AI infrastructure expands to diverse semiconductors and computing environments, software that integrates and operates them efficiently is becoming more important. Through SC26, where various AI semiconductor technologies come together, we will promote Backend.AI's scalability and integrated operations strategy, and actively respond to the AI infrastructure market, which is evolving around token production and operating efficiency."
Original Korean article: ZDNet Korea
English brief by Korea Wind Intelligence, koreawindintel.com















