NVIDIA turned its AI Day Seoul event into a showcase for Korea’s sovereign AI ambitions on 9 November, announcing a package that runs from chips to factory floors. With Korea’s science ministry behind it, the company will link Rebellions’ Korean-designed R200 processors to its NVLink Fusion platform, release deployment recipes for four Korean open models, and deepen physical AI work with Samsung, Hyundai, LG and Kakao Mobility. The message is blunt: NVIDIA wants Korea to build its own intelligence on NVIDIA’s platform, and it is opening that platform to Korean silicon to do it.
## The sovereign AI core: Korean chips on NVIDIA racks
The centrepiece is a collaboration between NVIDIA, Rebellions and SK Telecom, backed by Korea’s Ministry of Science and ICT. Rebellions will connect its next-generation R200 neural processing units to NVIDIA’s platform through NVLink Fusion, using the MGX rack architecture for rack-scale inference systems.
SK Telecom plans to deploy those systems in its AI data centres, tying them to its A.X K2 model and the government’s AI for All programme. The partners will split inference work across processors, using NVIDIA GPUs for the prefill stage and Rebellions NPUs for decode, aiming for faster and more power-efficient responses.
Rebellions’ R200 systems are expected to be ready for deployment in 2028, and the R200 itself is still in development. That timeline matters: this is a collaboration to build infrastructure, not a product shipping today, and the companies have disclosed no prices or order volumes.
## What ships today: recipes and a dataset
The concrete releases landed in software. NVIDIA is publishing Dynamo-optimised deployment recipes for four Korean open models — A.X K2 from SK Telecom, K-EXAONE 2.0 from LG AI Research, Solar Open 2 from Upstage and Motif-3 from Motif Technologies — tuned for its B200 GPUs and claiming up to twice the performance out of the box.
NVIDIA also released Nemotron-Personas-Korea, an open dataset of seven million fully synthetic Korean personas grounded in official Korean statistics, for builders training and testing models in Korean contexts. [LG’s K-EXAONE 2.5 API opening](https://geroundtable.com/lg-k-exaone-2-5-free-api/), announced at the same event, gives developers a free way to test one of those models now.
## Physical AI moves into the factory
The factory floor got the most measurable claims. Samsung is expanding its work with NVIDIA on accelerated computing and AI physics, and in one materials case a GPU-accelerated workflow ran roughly 1,600 times faster than the CPU version — cutting a simulation from about 18 hours to 41 seconds.
Kakao Mobility plans to build a level 4 autonomous driving system on NVIDIA’s Hyperion platform and DRIVE AGX Thor, with safety validation through NVIDIA Halos, starting with Kia purpose-built vehicles. LG Electronics demonstrated Isaac GR00T-powered robot work on its CLOiD machine in a real manufacturing setting, [building on earlier LG robot work with NVIDIA](https://geroundtable.com/lg-robot-data-fivefold-nvidia/).
## Why It Matters
Korea is the clearest test yet of NVIDIA’s sovereign AI pitch: a country with its own chips, models and factories, building intelligence for its own language and industries on NVIDIA’s platform. The NVLink Fusion opening is the structural shift — for the first time, a Korean chipmaker can plug its processors directly into NVIDIA’s rack-scale infrastructure.
The limits are just as clear. The R200 racks arrive in 2028, Hyundai is only exploring the agent tools, and much of the event re-presented October announcements such as Switchyard and OpenShell in a Korean setting. What is new today is the package: a government-backed, country-wide commitment rather than a single product launch.
## Our Read
NVIDIA’s play here is lock-in through openness. Let Korean chips plug into its racks, let Korean models run on its inference stack, and the country’s AI ends up routed through NVIDIA hardware even where the silicon is Korean. That is a smarter bet than selling GPUs alone, and it is one rivals cannot easily copy without the same rack-scale platform.
The honest test is delivery. The Dynamo recipes and the personas dataset are real and usable now, while the racks, the robots and the self-driving plans are promises with dates attached — and the gap between the two is where this story will be judged.
If the 2028 hardware lands on time, Korea becomes the template for every country that wants its own AI. NVIDIA would then be the company that sells them the shelves to put it on — [Mistral’s open-weight release](https://geroundtable.com/mistral-large-4-public-open-weight-release/) shows the other route, where openness means giving the weights away rather than the rack.
## Background
AI Day Seoul runs for two days at COEX in Seoul, with more than 2,000 developers, startups and researchers attending. NVIDIA has been pushing sovereign AI — national systems built on local language, data and infrastructure — as a core strategy, and Korea, with its chipmakers and manufacturers, is the most complete version of that pitch so far.
Source: [NVIDIA Blog](https://blogs.nvidia.com/ai-day-seoul-2026/) and [NVIDIA Newsroom](https://nvidianews.nvidia.com/news/nvidia-rebellions-and-sk-telecom-announce-collaboration-to-advance-sovereign-ai-infrastructure-with-nvidia-nvlink-fusion)
0 Comments