Korea’s AI hardware sector took the spotlight at the launch of OpenAI Korea, where Seoul-based AI unicorn FuriosaAI demonstrated its second-generation chip RNGD (Renegade) powering OpenAI’s open-source model GPT-OSS 120B. The event drew over 300 AI industry stakeholders and marked a rare instance of a Korean semiconductor startup proving its hardware against one of the world’s most demanding large language models. The showcase underscores Korea’s growing relevance in the global AI infrastructure race, where efficiency has become as critical as performance.

FuriosaAI Demonstrates RNGD at OpenAI Korea Launch

At the September 11 opening ceremony of OpenAI Korea, FuriosaAI was the only AI infrastructure company invited to demonstrate. Using just two RNGD chips, the startup successfully powered a real-time chatbot based on GPT-OSS 120B, OpenAI’s open-source foundation model with 120 billion parameters.

The GPT-OSS 120B model applies a Mixture-of-Experts (MoE) architecture, designed to maximize both performance and efficiency. FuriosaAI highlighted that running such a model with minimal hardware showed not only RNGD’s computational capability but also its power efficiency, a key bottleneck in scaling ultra-large models.

The Bottlenecks of Power and Cost in Global AI

The demonstration comes at a time when AI infrastructure faces mounting scrutiny over its escalating energy demands and operating costs. Training and running large language models (LLMs) such as GPT-4 or GPT-OSS 120B requires vast computational resources, raising concerns among policymakers and enterprises alike.

FuriosaAI’s approach directly addresses these challenges. By cutting the number of chips required and optimizing power-to-performance ratios, solutions like RNGD could reshape how startups, enterprises, and governments deploy LLMs. For Korea, the showcase reinforces national ambitions to position itself as a hub for AI semiconductor innovation, complementing broader ecosystem programs that support deep-tech scaleups and global expansion.

Strong Performance and Power Efficiency: Making AI More Sustainable and Accessible

Kim Han-joon, CTO of FuriosaAI, explained during the demonstration:

“With just two RNGD cards, we proved that a 120-billion-parameter model can be run in real time with strong performance and power efficiency.”

Baek Joon-ho, CEO of FuriosaAI, added:

“OpenAI’s mission — ensuring AGI benefits all of humanity — aligns with our own mission of making AI sustainable and accessible. The combination of RNGD and GPT-OSS will accelerate the growth of the global open-source AI ecosystem.”

South Korea as a Rising Player in Global AI Supply Chain

The RNGD demonstration is more than a product showcase; it signals how Korean startups are positioning themselves in the global AI supply chain. With U.S. and Chinese players dominating LLM software, Korea’s competitive edge increasingly lies in semiconductors and infrastructure efficiency.

By proving real-time operation of GPT-OSS 120B with reduced hardware, FuriosaAI shows a pathway to lowering total cost of ownership (TCO) for AI service providers. This matters for cloud operators, startups seeking to scale, and governments investing in sovereign AI capabilities.

FuriosaAI’s positioning also aligns with broader policy directions for the Korean ecosystem, such as the Ministry of SMEs and Startups’ focus on deep-tech competitiveness and the inclusion of AI semiconductors in Korea’s “growth engine” strategy.

Solving AI’s Vital Bottlenecks: Cost and Power Efficiency

At the end of the day, FuriosaAI’s RNGD demonstration at the OpenAI Korea launch highlights how Korean startups can contribute to solving AI’s most pressing bottlenecks: cost and power efficiency. As LLM deployment accelerates worldwide, efficiency breakthroughs will determine not just competitiveness but also the sustainability of AI adoption.

Global investors and founders can see in this case how Korea is emerging as a strategic testing ground for AI infrastructure innovation. Policymakers, meanwhile, gain a reminder that supporting deep-tech ventures is essential to securing long-term positions in the global AI economy.

– Stay Ahead in Korea’s Startup Scene –
Get real-time insights, funding updates, and policy shifts shaping Korea’s innovation ecosystem.
➡️ Follow KoreaTechDesk on LinkedIn, X (Twitter), Threads, Bluesky, Telegram, Facebook, and WhatsApp Channel.