Nvidia Takes Equity Stake in One Korean AI Startup, Plans Three More

At a government-hosted forum in Seoul on November 9, Nvidia announced a direct equity investment in one Korean startup and said it plans to invest in three others — moves that tie the chipmaker more closely to South Korea's push for "sovereign AI" across models, robotics and infrastructure.

The announcements came at the MSIT-NVIDIA AI Startup Partnership Forum, held at COEX in Seoul and co-hosted by the Ministry of Science and ICT (MSIT) and Nvidia. Ministry officials and Nvidia executives met with more than 20 Korean startups working on foundation models, physical AI, AI infrastructure and AI semiconductors [4][10]. Sources differ on which ministry official attended: one identifies Vice Minister Ryu Je-myung [4], while others name Lee Do-kyu, head of the Information and Communication Policy Office [3][7].

One completed deal, three planned

Nvidia said it has made a direct investment in Point2 Technology, a startup developing radio-frequency (RF) interconnects that move data between chips and racks in what Nvidia calls "AI factories" [10]. Point2, founded by Bae Hyeon-min, head of the KAIST startup institute, had already received funding from Nvidia's venture arm, NVentures, in the first half of 2026 [5][6].

For the other three companies, Nvidia used the phrase "planned investments" for all four in its blog post [10] — though other reporting distinguishes Point2 as a completed direct investment while the other three remain intentions [2]. No amounts, valuations, ownership stakes or timelines were disclosed for any of the four deals [2][3][4][5].

The three planned investments target:

• Upstage — a Korean enterprise AI company building the "Solar" series of large language models and document AI tools that extract information from complex business documents [10]. Upstage closed a Series C first tranche of 180 billion won (~$133.6 million) in April at a valuation above 1 trillion won (~$742 million), becoming Korea's first generative AI unicorn [3][4]. Amazon and AMD already hold stakes; Nvidia would join in the ongoing second tranche [3].

• Motif Technologies — a Seoul-based AI lab developing foundation models for AI agents, including Motif-3, released with open weights in August [3][7]. The company raised 24 billion won (~$17.8 million) in a Series B round in May and said it has closed a bridge round with Nvidia participation, though another report described Nvidia's investment in the bridge round as expected [3][7]. Motif plans to release Motif-4 in the first quarter of 2027 [3][7].

• RLWRLD (also rendered RealWorld) — a physical AI startup based in San Francisco, Seoul and Tokyo building robotics foundation models for humanoid and industrial robots [10]. Its models target five-finger dexterity tasks such as inserting components, threading nuts onto bolts and kitting parts from moving conveyors [2][10]. The company is reportedly seeking 200 billion won at a 1 trillion won pre-money valuation [4].

Stack-spanning strategy

The four startups span the AI stack: Point2 in interconnect infrastructure, Upstage and Motif in language models, and RLWRLD in robotics. Nvidia's blog described the group as "represent work across the full AI stack, from infrastructure to models to robotics applications" [10].

The same day, Nvidia signed a memorandum of understanding with SK Telecom and Rebellions to integrate Rebellions' next-generation NPU, the R200, into Nvidia's NVLink Fusion interconnect ecosystem — a separate but parallel move to bring Korean silicon into Nvidia's rack-scale architecture [3][8].

Sovereign AI context

The forum was explicitly framed as part of Korea's "sovereign AI" drive. Nvidia's blog said the collaboration "lays the groundwork for Korean NPU and XPU companies to explore participation in the NVIDIA NVLink Fusion ecosystem" and includes training, certification, hackathons and work with four Korean foundation-model builders using Nvidia Dynamo software [8]. The Ministry of Science and ICT said the project aims to create a benchmark for domestic AI data center construction and broaden cooperation to other Korean NPU companies [3].

Nvidia also noted that its Inception startup program already counts more than 1,000 Korea-based members, which receive preferred pricing, technical expertise and investor introductions [10][4].

What remains unknown

• Investment sizes: None of the four deals included a dollar figure, valuation, round structure or ownership percentage [2][3][4][5][6].
• Closing status: The three planned investments have not been completed and investment procedures have not been finalized [5][3]. Plans can change.
• Technical proof points: For Point2's RF interconnects, Nvidia provided no transmission distances, data rates, power figures, pricing or customer names [2][6]. The approach is presented as an alternative to copper and optical links, but no public benchmarks exist [2][6].
• Product integration: Nothing in the announcements ties the Point2 stake to future Nvidia products or to Korean memory suppliers [2].

Nvidia's framing

An Nvidia official stated that the company has supported more than 1,000 domestic startups through its Inception startup growth support program, and that this investment plan is being pursued based on the cooperation Nvidia has maintained with the domestic startup and venture ecosystem [4].

Nvidia's blog post announces all four companies under the header "planned investments" [10] — a distinction that some headlines have blurred [2].

Why it matters

Nvidia is using equity to lock in early relationships with startups that could become significant consumers of its hardware and software across language modeling, robotics and interconnect technology. The government co-hosting gives the announcements policy weight, aligning with Seoul's goal of moving Korean firms up the AI value chain beyond memory supply. But until terms are disclosed and the planned investments close, the scale of Nvidia's commitment — and the commercial traction of the startups involved — remains an open question.