Test report DSG-1732 · Rev A · tested November 17, 2026
AI Datacenter InfrastructureDevice under test
Nvidia to Build 1 Million Korean-Language Data Sets by Next Year
Nvidia will build over 1 million Korean-language data sets by next year, opening results to Korean AI firms with commercial use permitted under its $300 million KAIST center.
- Read
- 3 min
- Words
- 543
- Node
- 10nm
- Operator
- Marcus Bennett
Spec summary
- Nvidia plans more than 1 million Korean-language data sets by next year.
- The Nvidia-KAIST center invests $300 million over five years, announced in July.
- $250 million of the budget goes to computing resources at $50 million per year.
- Research results will support Nvidia's Nemotron and Korean proprietary models, with commercial use permitted.
- First project results are expected around the second half of next year.

Nvidia will build more than 1 million Korean-language data sets by next year, according to the head of its joint research center with KAIST. Kim Hyun-woo, who leads the Nvidia-KAIST Joint AI Research Center, announced the plan at a press briefing on the 10th at Nvidia Korea in Seoul's Gangnam district.
"Data matters as much as GPUs," Kim said. "It is genuinely difficult to find publicly available, high-quality Korean-language data sets."
The center plans to build the data sets at the document level. Timing and method of release will be decided as the research progresses.
What does this change for Korean AI developers?
The center's orientation is the notable part. It will not only build AI models for Nvidia. It plans to provide foundational technology that Korean companies can use to develop AI more effectively. Research results will be released for use both in Nvidia's own model, Nemotron, and in proprietary models built by Korean firms. Commercial use will be permitted.
The center will also jointly develop synthetic data technology, where AI generates new training material. It will work on methods for learning more information from smaller volumes of data.
Kim likened the approach to trees and a forest. "The goal is not to grow one good tree and transplant it, but to study how to create a forest and share that with the world," he said. The center will focus less on improving one specific model and more on laying a technical foundation that multiple companies can use.
How is the $300 million split?
Nvidia and KAIST will invest a total of $300 million, about 400 billion won, over five years in the joint center, which was announced in July. More than 80% of the total, $250 million, is allocated to computing resources. That works out to $50 million a year over five years. The remaining $50 million covers personnel and operating costs.
The center is still securing the GPUs it needs. Procurement will run through Nvidia's cloud partners in South Korea. Specific models, quantities and timing remain undetermined. "Because the volume is large, we are in discussions to procure the supply," Kim said.
Who is being hired?
Recruitment is accelerating. The center plans to recruit several full-time researchers employed by Nvidia this year and expand its infrastructure in stages.
A separate program, distinct from full-time hiring, will support at least 10 KAIST researchers each year. It will also offer Nvidia internship opportunities.
When will results appear?
The outline of the center's projects should emerge as early as the second half of next year. "We have some interesting projects lined up, and I expect the first results will be unveiled around the second half of next year," Kim said.
Why the focus on data?
Kim said investment in AI models and computing infrastructure has expanded, but the accumulation and sharing of Korean-language data has lagged. High-quality training data sits at the core of AI competitiveness, and Nvidia is stepping directly into securing it for the Korean market.
The move broadens the domestic AI industry by addressing a scarcity of public, high-quality Korean-language material — a gap that individual model developers have struggled to close on their own.
via en.sedaily.com (Source)
More from Marcus Bennett
Show full bio
News editor covering regulation, policy and their impact on hardware layer of artificial intelligence at Die Signal.
681 articles
Same lot · LOT-362D
- DSG-754228nmSeoul Daily Calls for K-NPU Roadmap to Enter AI Top Three
- DSG-784628nmKorea Convenes Emergency GPU Supply Review After Nvidia CEO Exit
- DSG-644620nmNvidia, Rebellions, SK Telecom target 2028 AI-chip deployment
- DSG-796014nmNvidia’s Seoul AI Day: 260,000 GPUs, Huang Investigation
- DSG-863110nmJensen Huang Exit Risk Splits Korean AI Startups on Stakes