• Sharing Open AI Infrastructure Leadership at Korea’s Largest Open Data Center Technology Conference

Lablup, an AI infrastructure company led by CEO Jeongkyu Shin, announced that it will participate in the 2026 OCP Korea Tech Day, to be held on August 21 at the COEX Grand Ballroom in Samseong-dong, Seoul. The company will present its work on confidential-computing-based security for AI agents and the development and operation of a heterogeneous AI accelerator testbed. Lablup will also operate an exhibition booth at Booth 13 throughout the event.

OCP Korea Tech Day is Korea’s largest open data center technology conference, hosted by the Open Compute Project (OCP) Foundation and the OCP Korea Community since 2019. This year’s event will cover core technologies for AI data centers, from servers, storage, networking, and cooling to AI clusters and next-generation silicon packaging. The program is expected to bring together approximately 1,500 participants from industry, academia, and research institutions, with around 60 keynote and technical sessions and more than 30 exhibition booths.

Lablup will share its expertise in operating open AI infrastructure through two technical sessions.

Joongi Kim, Chief Technology Officer at Lablup, will present “Enhancing AI Agent Security and Privacy with Confidential Computing.” The session will explore how confidential computing, combined with AI accelerators, can enable the deployment and operation of AI agents that safely process sensitive and personal data, even on AI infrastructure operated by third-party providers. Kim will discuss recent technical developments and system-design considerations for building confidential AI agents.

Yubin Kim, Software Engineer at Lablup, will present “Building and Operating an AI Accelerator Testbed by Integrating OpenStack-Based Private Infrastructure with Public Clouds.” As heterogeneous AI accelerators continue to proliferate, infrastructure validation has become a growing operational bottleneck. Public clouds, closed virtualization platforms, and directly managed bare-metal environments each introduce distinct constraints, including limited accelerator availability, vendor lock-in, and inefficiencies caused by manual operations.

The session will introduce Lablup’s multi-provider abstraction testbed, which integrates an in-house OpenStack environment with public clouds such as AWS and Microsoft Azure through a unified API. The architecture enables flexible validation across heterogeneous hardware environments. It will cover practical patterns that organizations can apply immediately, including automated provisioning of on-premises nodes and standardized integration of emerging architectures such as NVIDIA DGX Spark, which features a unified memory architecture.

In addition to its sessions, Lablup will showcase AI infrastructure solutions spanning desktops to data centers at its exhibition booth, centered on its open-source AI infrastructure platform, Backend.AI. Visitors will be able to learn more about Lablup’s core technologies, including unified operations for heterogeneous accelerators and container-level GPU fractional virtualization.

“Achieving sustainable growth in AI infrastructure depends on open standards and collaboration, rather than dependence on a single vendor,” said Joongi Kim, CTO of Lablup. “We hope that Lablup’s accumulated experience in infrastructure operations and security technologies for the heterogeneous accelerator era will contribute to Korea’s data center ecosystem.”