NVIDIA backs Point2 and plans three Korean AI startup investments

NVIDIA announced an investment in Point2 and three planned Korean AI startup investments on November 9, 2026: Motif Technologies, RLWRLD and Upstage. The slate extends NVIDIA’s Korean startup activity across AI data-center interconnects, open-weight language models, robotics and enterprise document software.

The stakes are deployment coverage across the AI stack. Point2 works on the links between chips and racks; Motif builds models for agents; RLWRLD builds models for robots; and Upstage packages AI for document-heavy business workflows. The announcement ties NVIDIA’s Korean capital to both infrastructure and applications.

What did the November 9 forum add?

The announcement accompanied the MSIT-NVIDIA AI Startup Partnership Forum at COEX, hosted by Korea’s Ministry of Science and ICT and NVIDIA.

Vice Minister of Science and ICT Ryu Je-myung joined NVIDIA leaders to meet more than 20 Korean startups working on foundation models, physical AI, AI infrastructure and AI semiconductors.

The four named companies add a direct investment component to those broader startup meetings. Their roles differ by technical layer:

• Point2 Technology: Radio-frequency interconnects for moving data between chips and racks in AI data centers.
• Motif Technologies: Open-weight foundation models for AI agents, including Motif 3.
• RLWRLD: Robotics foundation models for humanoid and industrial robots.
• Upstage: Solar language models and document-AI products for enterprise workflows.

The forum’s wider agenda places the capital inside a government-industry channel for Korean AI development. The more than 20 startups covered foundation models, physical AI, infrastructure and semiconductors, a broader group than the four direct-investment names.

Why is Point2 part of NVIDIA’s AI cluster strategy?

Point2 Technology develops radio-frequency interconnects that move data between chips and racks in AI data centers. Its e-Tube documentation describes a “purely electrical design” that sends millimeter-wave radio-frequency signals through a plastic core called a dielectric waveguide.

Point2 says the design avoids the conversion between optical and electrical signals required by optical cabling. That puts a radio-frequency link against the optical alternative used in AI clusters and gives NVIDIA’s infrastructure investment a specific hardware target: the links needed to scale AI clusters.

NVIDIA’s relationship with Point2 was already visible before the November announcement. On August 10, 2026, Point2 announced a financing extension that brought its total Series B financing to $136 million and listed NVIDIA among the backers. The earlier raise shows that the chipmaker had backed the company before the new investment.

What do Motif 3 and RLWRLD bring to the slate?

Motif Technologies, a Seoul-based AI lab, develops foundation models for building AI agents. Its model page lists the open-weight Motif 3 under an MIT license and presents the model for reasoning, coding and tool use in multi-step tasks. The investment adds a model developer to a slate that otherwise concentrates on infrastructure, physical machines and workflow software.

According to Motif’s technical report, submitted on August 10, 2026, Motif 3 uses a mixture-of-experts design that activates only part of the model for each token. The report lists 314 billion total parameters, with 13.2 billion parameters activated per token.

The report also records approximately 12.5 trillion pretraining tokens and a context window of up to 256,000 tokens. Motif’s report links that window to the amount of material the model can consider at once during long-document or multi-step agent tasks. The figures distinguish the model’s total capacity from the portion activated for each token.

RLWRLD develops robotics foundation models for humanoid and industrial robots performing work that requires five-finger dexterity. The company lists inserting components, threading nuts onto bolts and collecting parts from moving conveyors as target tasks. The startup operates from San Francisco, Seoul and Tokyo.

The RLDX-1 technical report, first submitted on May 5, 2026, describes a Multi-Stream Action Transformer. It integrates different modalities through dedicated streams for motion awareness, long-term memory and physical sensing. RLWRLD says the design addresses those areas rather than relying only on vision and language.

The named streams extend the model beyond a vision-language interface. They target the sensing, memory and motion information that RLWRLD says humanoid and industrial robots need for the dexterity tasks it lists.

How does Upstage fit enterprise deployments?

Upstage brings enterprise workflow software into the slate through its Solar language models and document-AI products. Its product site says Document Parse converts PDFs, scans and emails into machine-readable text. Information Extract pulls structured data from invoices, claims and contracts.

Upstage lists three deployment routes: a console with REST API access, AWS Marketplace and on-premises installation behind an organization’s own firewall. The company presents the on-premises option as support for data sovereignty and compliance.

The firewall deployment keeps document processing within an organization’s infrastructure. Upstage pairs that route with software distribution through its own console and AWS Marketplace, giving enterprises three ways to deploy the same document-processing products.

What does Inception provide beyond capital?

Beyond the four named investments, NVIDIA says its Korean startup network includes “more than 1,000 Korea-based Inception members.” Inception is free and does not require equity. Its benefits include technical training and developer tools, preferred pricing on select NVIDIA hardware and software, and cloud credits from NVIDIA and partners.

Eligible startups can also gain investor access and curated networking events. The program differs from NVentures, NVIDIA’s venture-capital arm, which focuses on strategic investments. NVIDIA therefore presents Inception as an ecosystem-wide service rather than a portfolio of direct venture positions.

The membership figure describes reach across the ecosystem, not capital deployed. Inception provides broad support without requiring members to surrender equity, while the four announcements identify selective capital for specific technical layers. The two structures address different scales of need and remain distinct programs.

What did the January meeting set up?

The ministry’s startup coordination with NVIDIA predates the November forum. At a January 9, 2026 meeting, Ryu Je-myung and NVIDIA executive vice president Jay Puri discussed linking Korea’s K-Digital Grand Championship with NVIDIA’s Inception Startup Grand Challenge. The ministry describes the former as a government initiative for early-stage AI startups.

The January discussion connected two startup-support channels. The November forum adds a named investment component to that earlier effort and pairs it with a ministry-hosted meeting of more than 20 Korean startups. Together, the programs combine free ecosystem assistance with selective capital across the AI stack.

With Point2 announced and the other three companies still marked as planned, completion of the Motif Technologies, RLWRLD and Upstage investments is NVIDIA’s next measurable step in Korea.