"Token Costs Are Key to an AI-Driven Society... Domestic NPUs Essential"

Renegade 2026 Summit Hosts Tokenomics Panel Discussion

"Token Costs Will Determine AI Competitiveness"

"Increased AI Usage Leads to Higher Cost Burdens"

"Affordable Infrastructure Needed with Domestic NPU Technology"

"Expanding Public Intelligence with AI Requires Low-Cost Solutions for Wider Social Impact"

On the 2nd, a discussion on "Sovereign AI Advancement and the Post-Era NPU-Based AI Token Economy" is underway during the "Renegade 2026 Summit" event. From the left, Jongho Paek, CEO of Furiosa AI; Jiho Cha, Member of the National Assembly; Woohyung Lim, Head of LG AI Research Institute. Photo by Jongmin Paek, Tech Specialist

There is a growing opinion that in order to implement 'Tokenomics'—an economic structure discussed as a key component of the artificial intelligence (AI) era—at a lower cost, the current GPU-centric ecosystem must be transitioned to a domestically produced Neural Processing Unit (NPU)-centered system. The importance of NPUs was also emphasized as the infrastructure foundation for the government's policy initiatives toward an AI-driven society.

On April 2, at the Renegade 2026 Summit hosted by Furiosa AI, an NPU company, at SJ Kunsthalle in Gangnam, Seoul, a panel discussion was held under the theme "AI Tokenomics on NPU in Sovereign Development & Beyond." Moderated by Junho Paek, CEO of Furiosa AI, the discussion featured Sung Kim, CEO of Upstage; Woohyung Lim, Head of LG AI Research; and Jiho Cha, Member of the National Assembly from the Democratic Party of Korea. This brought together semiconductor industry leaders, AI researchers, and lawmakers to discuss AI cost structures and public sector utilization in one place.

Tokenomics is an economic concept centered on the production and consumption of 'tokens' generated in the process of utilizing AI services, as well as their associated cost structures. It has recently emerged as a major topic of discussion amid the rapid advancement of AI. In this context, a token refers to the smallest unit of data used by AI when processing text, images, and other content according to user instructions. AI companies typically charge users based on the amount of tokens consumed.

Following the emergence of OpenClo and the rapid rise of agentic AI, the burden of token costs has grown even more significant. An increase in token expenses not only places a heavier burden on AI users but also becomes a key variable affecting the speed at which services can proliferate. While service sophistication increases user convenience, it simultaneously leads to higher token costs. For this reason, participants repeatedly emphasized the importance of cost structures and building affordable infrastructure, considering them just as critical as AI model competition itself.

During the discussion, Assemblyman Jiho Cha linked the topic of tokenomics to issues of the public sector and global order. As a key figure leading AI-driven societal policies, he explained that if AI is developed in the form of 'public intelligence,' it could become universally accessible to everyone, and could be further expanded into various social sectors such as education, healthcare, finance, and welfare.

Assemblyman Cha stated, "We should ask ourselves whether we will continue to pursue the highest performing chips, or whether chips that deliver 95% of the performance at a much lower and more efficient cost will dominate the global market."

This suggests that for public intelligence to function in practice, large-scale users must be able to access services simultaneously, and the system must have the token processing capability to support such demand.

Assemblyman Cha went on to say, "The important issue is not competition, but who sets the rules." This was interpreted as highlighting the potential for Korea to play a leading role in shaping a new AI order—not only by developing independent models, but also by establishing a robust semiconductor infrastructure.

Sung Kim, CEO of Upstage, who is developing a proprietary Korean foundation AI model, said, "In the coming AI era, the most important factors will be how much tokens are consumed and how much are generated," adding, "To achieve comprehensive tokenomics, demand, service, model, and infrastructure are all needed, and among these, the most critical is an NPU that can run models at lower cost." He argued that NPUs are the core infrastructure for reducing token expenses.

Woohyung Lim, Head of LG AI Research, who utilizes Furiosa's NPUs, also commented, "The biggest bottleneck in using AI is hardware infrastructure," adding, "Chips such as NPUs will play a particularly important role." He asserted that as AI demand expands into practical applications, not only data centers but also the semiconductor infrastructure that underpins them will become increasingly important.

Before the panel began, Jaemyoung Ryu, Second Vice Minister of Science and ICT, sent a congratulatory video message, stating, "The Ministry of Science and ICT has provided active support for Furiosa AI since 2020. The company is one to watch in the global AI race." He reaffirmed the government’s commitment to supporting the development and adoption of domestic NPUs.

▲Token: The smallest unit of data used by AI when processing text, images, etc. Costs are calculated based on the amount of tokens consumed.

▲AI-Driven Society: A societal model in which AI is utilized as public infrastructure to provide accessible intelligent services in sectors such as education, healthcare, and welfare for all.

This content was produced with the assistance of AI translation services.