DS&G Backs Korea University's InThon Hackathon With Nvidia B300 Servers

South Korean AI infrastructure company DS&G is sponsoring the 2026 InThon hackathon and datathon at Korea University with two Nvidia HGX B300 8-GPU servers. The event runs from October 31 to November 1 and is organized by the university's College of Informatics, its student council and the AI-centered university program.

Now in its fourth year, InThon draws about 150 students majoring in software and AI. Hackathon teams have 24 hours, with no overnight break, to build a working app or web service on the given theme. DS&G will provide the B300 servers through remote access. They will serve as core compute for large language model (LLM) inference and AI service operation.

Lablup joins as co-sponsor

AI infrastructure software company Lablup is sponsoring the hackathon track and giving participating teams vouchers for its Lablup Token Factory, an LLM API service. Students will connect the Token Factory API to coding agents to develop a range of AI-based applications.

DS&G and Lablup also plan to collect anonymized traces of the timing and size of coding-agent requests during the event. Using them, they will compare throughput and response latency under different serving configurations on the Nvidia B300. The results will be published on Lablup's technical blog and are meant to serve as reference material for companies building and running AI infrastructure.

Company background

DS&G is an Elite partner in the Nvidia Partner Network, the program's top tier above Registered and Preferred, and the largest Supermicro distributor in South Korea. Its executive director Kang Shin-bong gave a technical session on October 27 at a pre-event team-building session, covering AI Factory infrastructure and the shift from generative AI to agentic and physical AI.

Lablup CEO Jeong-kyu Shin said the B300 servers let the company examine serving optimization for coding-agent workloads. DS&G executive director Seo Young-min said the sponsorship helps show how B300-based infrastructure performs in real coding-agent environments and will inform its support for enterprise AI infrastructure.