Korea and Europe Are Buying National AI Stacks, Not Just Models

Korea is buying the whole stack

NVIDIA and Korea's Ministry of Science and ICT (MSIT) are expanding their collaboration across the country's sovereign AI ecosystem, and the scope is the story. The work covers AI semiconductors, data centre infrastructure, talent, startups, models, agents and security. That is not a model deal or a chip sale. It is a national stack, assembled in public, with the government ministry as a partner. The collaboration also lays the groundwork for Korean NPU and XPU companies to explore participation in the NVIDIA NVLink Fusion ecosystem, which would put domestic chipmakers inside the interconnect layer instead of outside it. m:…8955

The same collaboration reaches into software and skills. NVIDIA is working with four leading Korean foundation model builders, using its Dynamo software to optimise and scale AI inference, and with three selected consortia under Korea's AI for All initiative. A separate collaboration with SK Telecom and Rebellions, supported by MSIT, aims to advance sovereign AI infrastructure for local models and workloads. NVIDIA is also providing training and certification programmes, hackathons and technical seminars to strengthen GPU use in Korea. Hardware, models, people and deployment, all at once. m:…8955

Startups fill in the layers

MSIT and NVIDIA hosted the MSIT-NVIDIA AI Startup Partnership Forum at COEX, where Ryu Je-myung, vice minister of science and ICT, met with more than 20 Korean startups across foundation models, physical AI, AI infrastructure and AI semiconductors. During the event NVIDIA announced an investment in Point2 Technology, with further investments planned in Motif Technologies, RLWRLD and Upstage. NVIDIA Inception, which supports startups with preferred pricing, technical expertise and investor introductions, has more than 1,000 Korea-based members. m:…3716

The four named startups cover different layers of the stack. Point2 Technology builds scalable, radio frequency-based interconnects that move data between chips and racks in AI factories. Motif Technologies, a Seoul-based AI lab, develops foundation models for building AI agents and has released Motif-3, a large language model with open weights. RLWRLD, a physical AI startup based in San Francisco, Seoul and Tokyo, develops robotics foundation models. Upstage, a Korean enterprise AI company, develops Solar language models and document AI tools. m:…3716

Europe's answer is vertical integration

In Europe, Mistral CEO Arthur Mensch told the 20VC podcast that sovereign AI is the most critical category for European economic independence, warning that reliance on centralised US providers creates long-term trade balance and security risks. He said, "sovereign AI is going to be the most important category." His answer is vertical integration: training models, running infrastructure and building custom business applications, so companies keep control over their data and intellectual property instead of renting it back from a distant provider. m:…2005

Mistral focuses on four verticals: manufacturing, financial services, defence and the public sector. The company says about 60% of its revenue base is European and about 80% of its token traffic comes from its own models. It is scaling customer-facing engineering through partnerships with system integrators such as Accenture and TCS, and it does not plan to build its own AI chips. Mensch puts energy at around 10% of the total cost of owning compute, and says the real bottleneck is compute and capital, not energy. He also says the open-weight frontier is catching up with the closed frontier, and in many domains is above it. m:…2005

India's version: ownership inside the enterprise

India's Sarvam AI is building its version of the same idea from the enterprise side. It has appointed Sashikumar Sreedharan as President and Chief Business Officer and a member of its board. Sreedharan joins from Google Cloud, where he most recently served as Managing Director for India, and previously led Microsoft India's enterprise business as Managing Director for over five years. He will lead Sarvam's business across enterprises and government. The company says more than 100 enterprises and government institutions are already building with its full-stack offering, and its position is that every enterprise should own its AI: the data, the models and the agents stay inside the enterprise, and so does the value they create. m:…7261

The stack now reaches the factory floor

In Korea the stack does not stop at models. Samsung Electronics is expanding its collaboration with NVIDIA to advance semiconductor innovation through accelerated computing and AI physics, using NVIDIA cuISS and cuDSS to accelerate computations involving sparse linear systems in technology computer-aided design. Its sMOL platform has run selected quantum chemical calculations up to five times faster end to end with NVIDIA cuEST, and in one AI homogenisation case, combining AI-predicted material properties, coarser grids and GPU acceleration produced a roughly 1,600x speedup, cutting simulation time from about 18 hours on a 32-core CPU to 41 seconds. m:…8953

Kakao Mobility plans to build a level 4 autonomous driving system on NVIDIA's autonomous vehicle computing and safety technologies, using NVIDIA Hyperion, powered by NVIDIA DRIVE AGX Thor, and NVIDIA Halos for safety across development and validation, with initial development planned around Kia purpose-built vehicles. NVIDIA and LG Electronics demonstrated NVIDIA Isaac GR00T-powered robotic manipulation and manufacturing tasks using LG's CLOiD robot in an actual manufacturing environment, with a data flywheel connecting real-world data collection, NVIDIA Cosmos-based augmentation, GR00T model training and simulation-based evaluation. Sovereign AI, in Korea at least, is now a factory programme as much as a model programme. m:…8953