NVIDIA and South Korea's largest companies disclosed a bundle of AI infrastructure plans at an AI Summit held in San Francisco in July 2026. NAVER, Brookfield and NVIDIA will expand NAVER's NVIDIA DSX AI factory at the GAK Sejong data center to 200 megawatts, roughly 100,000 GPUs. Hyundai Motor Group laid out a physical AI strategy that includes a buildout of 50,000 NVIDIA Blackwell GPUs, and SK Group and NVIDIA announced plans for a comprehensive partnership valued at more than $500 billion. ASAP works from NVIDIA's official blog as the primary source to lay out what was announced and how much of it is settled.

The infrastructure numbers disclosed at the summit

NAVER's expansion is the largest single item in the announcement by facility scale. NAVER, Brookfield and NVIDIA will expand NAVER's NVIDIA DSX AI factory at the GAK Sejong data center to 200 megawatts. NVIDIA describes that capacity as roughly 100,000 GPUs and names the NVIDIA Vera Rubin platform as the foundation.

SK Group put forward the larger headline figure. SK Group and NVIDIA announced plans for a comprehensive partnership of more than $500 billion, with the expanded collaboration deploying NVIDIA Vera Rubin infrastructure powered by SK hynix HBM4 memory. Ahead of the summit, CEO Jensen Huang, SK Group Chairman Chey Tae-won, and teams from SK hynix, SK Telecom and NVIDIA held a dinner in Woodside, California marking their longstanding memory codevelopment partnership.

What Hyundai Motor Group put under "physical AI"

Hyundai Motor Group's announcement points at machines that move rather than at data centers. The group led with a robot reference platform developed with NVIDIA and said it is extending a buildout with 50,000 NVIDIA Blackwell GPUs. On the autonomous driving axis, NVIDIA DRIVE Hyperion, an autonomous vehicle development platform, is being integrated with HMG vehicle platforms alongside NVIDIA DRIVE AGX in-vehicle computing and the safety-certified NVIDIA DriveOS operating system.

The university track: a KAIST joint lab and two NVAITCs

University collaboration is as notable in this bundle as the industrial infrastructure. NVIDIA and KAIST announced a joint AI research lab at the KAIST Kim Jaechul Graduate School of AI in Seoul, dedicated to advancing agentic AI for South Korea. NVIDIA describes it as the first joint lab between a Korean university and a global technology company, and the collaboration brings together NVIDIA's full-stack expertise, Nemotron open models and NVIDIA AI Cloud computing with KAIST's research talent.

The university track extends further. NVIDIA and KAIST's Department of Mechanical Engineering planned an NVIDIA AI Technology Center (NVAITC) for physical AI research, talent development and knowledge exchange using Nemotron open models. A separate NVAITC is planned with Seoul National University covering foundation models, accelerated computing, physical AI and AI for science, drawing on Nemotron and NVIDIA Cosmos open world foundation models.

The venue was San Francisco, not Seoul

The most revealing detail of the summit is the location rather than any individual number. President Jae Myung Lee and Korea's top business leaders announced Korea's AI infrastructure plans in San Francisco, and the panel seated Jensen Huang alongside OpenAI's Sam Altman, Broadcom's Hock Tan and Anthropic's Dario Amodei. The Korean side included Samsung Electronics Chairman Jay Y. Lee, SK Group Chairman Chey Tae-won, Hyundai Motor Group Executive Chair Euisun Chung and NAVER founder Hae-jin Lee, with Stanford's Soh Kim moderating.

Venue here is structure, not symbolism. It signals that the binding constraint on Korea's plans sits in GPU supply and partnership access rather than in domestic decision-making. When much of what a national plan needs depends on a foreign supplier's allocation queue, the contracts and the announcements happen where the supplier is. President Lee's "San Francisco AI Declaration," framed as a multitrillion-dollar push into AI semiconductors, hyperscale data centers and physical AI, reads the same way. Declaring a national program at the heart of the supply chain says plainly where the program's success is decided.

How to read the $500 billion figure

The "$500-billion-plus" number from SK Group and NVIDIA needs its impressive part separated from its cautious part. What is impressive is the order of magnitude. A figure at that scale belongs to the vocabulary of national industrial policy rather than to any single data center project, and it presumes a memory supplier and an accelerator supplier binding long-horizon volume together. It is also worth noting the unit: NAVER's 200 megawatts is a physical quantity and Hyundai's 50,000 is a countable one, while $500 billion is the most elastic unit of the three.

What deserves caution is that the announcement is written in the language of plans. NVIDIA's official post carries no disbursement schedule, no year-by-year allocation and no statement about how binding the commitment is. The number is therefore more accurately read as a ceiling and a direction than as committed spending. In infrastructure announcements, the larger the figure the longer the horizon, and long horizons absorb revisions. That is the point at which this number is most often misquoted.

Where the center of gravity moved

What this bundle documents is a shift in Korea's AI conversation from building data centers to applying them in industry. The evidence sits in the composition: Hyundai Motor Group's item combines GPU counts with a robot reference platform and vehicle platform integration, and physical AI appears explicitly in both the KAIST and the Seoul National University collaborations.

That shift fits the country's industrial structure. Competing at the frontier on models themselves is a harder path to demonstrate returns on than layering AI onto manufacturing, vehicles and memory, where position in world markets is already established. Huang's remark that without the invention of HBM memory in Korea NVIDIA would not have invented the supercomputers powering AI today, and Altman's that there really would not have been this AI revolution without Korea, both locate Korea's contribution in hardware and manufacturing rather than in models. For teams working in Korea, the implication is concrete: a large share of near-term AI demand is likely to surface in factory, vehicle and robot control and simulation rather than in chat interfaces.

Open questions: power, timing, and the distance between announcement and execution

What the announcement leaves unstated matters as much as what it states. NVIDIA's post does not say by when NAVER's 200 megawatts is secured, or where and how that power is sourced. Before 200 megawatts converts into a GPU count it is a generation, transmission and cooling problem, and those conditions move far more slowly than contracts do.

The distance between announcement and execution stays open too. The joint lab and both NVAITCs are described at the planning stage, and Hyundai Motor Group's autonomous driving collaboration is presented as a direction for platform integration. The verifiable moment in announcements of this kind is usually the first output, and what matters until then is not the dollar figure but what the deployed GPUs actually train and actually control. If this summit was where the direction of Korea's AI infrastructure was declared, the measure of whether that direction became results will be built the next time these parties report outcomes rather than numbers.

Source: ASAP summary based on NVIDIA's official blog post "At AI Summit, South Korea Outlines Its AI Future With NVIDIA and Partners" (July 24, 2026).