NVIDIA Expands Korean Startup Investments, Dynamo Recipes and Robotics

At the MSIT-NVIDIA AI Startup Partnership Forum held on November 9, 2026 at COEX in Seoul, NVIDIA announced new venture investments and technical collaborations alongside Korea's Ministry of Science and ICT (MSIT) and Vice Minister Ryu Je-myung. The gathering convened more than 20 Korean startups and built upon an ecosystem of over 1,000 Korean companies participating in the NVIDIA Inception program, as well as the N&UP startup acceleration initiative that MSIT and NVIDIA launched in 2020 and expanded in October 2025 when Korea joined the global NVIDIA AI Nation effort.

During the Seoul forum, NVIDIA confirmed a completed investment in Point2 Technology, a developer of radio-frequency-based interconnect solutions engineered to break bandwidth bottlenecks between racks and data centers while lowering power consumption and cost compared with optical cables. NVIDIA also outlined planned investments in 3 additional Korean foundation model and robotics companies: Motif Technologies, creator of the Motif-3 open-weights large language model; RLWRLD, a physical AI developer operating across San Francisco, Seoul, and Tokyo that builds five-finger robotics foundation models trained on real-world industrial factory datasets; and Upstage, creator of the Solar family of large language models and enterprise document intelligence software.

To accelerate deployment of sovereign foundation models across Korean data centers, NVIDIA simultaneously released open-source NVIDIA Dynamo deployment recipes optimized for NVIDIA B200 GPUs. Authored with contributions from Korean model builders, the Dynamo distributed inference framework delivers up to a 2x performance increase out of the box—corresponding to 2 times higher throughput across 4 open Korean foundation models—by coordinating TensorRT-LLM, vLLM, and SGLang engines with disaggregated prefill and decode scheduling and KV-cache-aware routing. The 4 supported models selected under MSIT's Sovereign AI Foundation Model Project are SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2, and Motif Technologies' Motif-3.

Across industrial manufacturing and autonomous mobility, NVIDIA expanded its Korean physical AI ecosystem collaborations with Samsung Electronics, Kakao Mobility, and LG Electronics. Samsung Electronics integrated the NVIDIA cuISS library for device-level semiconductor simulation, the cuDSS direct sparse solver library, and the cuEST quantum chemistry library—which delivered a 5x speedup (5 times faster execution) on Samsung's sMOL molecular simulation software—alongside cuEquivariance and the NVIDIA PhysicsNeMo framework, which reduced plasma-enhanced chemical vapor deposition simulation runtimes by 1,600x (1,600 times faster), cutting execution on a 32-core CPU from 18 hours down to 41 seconds on a single GPU. In mobility and home robotics, Kakao Mobility adopted NVIDIA DRIVE Hyperion, DRIVE AGX Thor, and the NVIDIA Halos safety system to deploy Level 4 autonomous driving on Kia purpose-built vehicles, while LG Electronics used NVIDIA Isaac GR00T, NVIDIA Cosmos world foundation models, and Jetson AGX Thor to power its CLOiD domestic robot.

In enterprise software and national infrastructure, NVIDIA detailed new secure agentic AI deployments across Korea as Hyundai Motor Group adopted NVIDIA NemoClaw blueprints and the OpenShell runtime for autonomous driving and manufacturing agents, security startup AIM Intelligence integrated NemoClaw and NeMo Guardrails into its Starfort red-teaming platform, and LG AI Research released its K-EXAONE-2.5 API trained with the open Nemotron-Personas-Korea dataset of 7 million synthetic Korean personas. Supporting those workloads at the hardware layer, NVIDIA's sovereign AI infrastructure collaboration with MSIT connects 3 industry consortia and the 4 foundation model builders with SK Telecom and Korean AI chipmaker Rebellions through NVIDIA NVLink Fusion high-speed interconnects.