Skip to content

From chip cables to fingertip dexterity: Nvidia backs four Korean firms to cover AI layers

Share
From chip cables to fingertip dexterity: Nvidia backs four Korean firms to cover AI layers

Listen to this article

Read by Anchor

Nvidia's investment in four Korean start-ups reflects an explicit shift in AI hardware priorities and foundational models. During the partnership forum organized by the Korean Ministry of Science and ICT together with Nvidia at the Quicks conference centre, Deputy Minister of Science Ryu Ji-myung met with leaders of the US company to discuss the future of more than twenty Korean firms specializing in foundational models, physical AI, infrastructure and semiconductors. The official announcement included a direct investment in Point2 Technology, alongside plans for additional investments in three other companies: Motif Technologies, RLWRLD and Upstage.

The real bottleneck in compute centres is no longer limited to processor speed but to the pathways that move data between chips and server racks.Here Nvidia's bet on Point2 stands out; the company is developing scalable interconnects that use radio frequencies to transmit data inside AI factories. This solution offers a unconventional alternative to copper cables and the usual optical solutions employed in scaling high-performance compute clusters, aiming to deliver the range, performance and efficiency required for the next generation of heavily utilized data centres.

At the software level, the investment map expands to cover the entire stack. In the heart of Seoul, Motif Technologies’ lab focuses on developing foundational models to build intelligent agents that perform highly complex tasks, equipped with its large language model Motif-3 released with open weights for developers. Along the same track, Korean Upstage leads the development of Solar language models and document-intelligence tools, solutions aimed at the enterprise sector that enable organisations to extract precise data from complex documents and automate daily workflow processes.

Moving from text generation to physical control represents the fourth pillar of this technical alliance.RLWRLD, which has offices in San Francisco, Seoul and Tokyo, offers foundational models specialized in physical AI and robotics. These models are designed to give humanoid robots and industrial systems a five-finger hand-like motor skill, allowing them to perform precise tasks such as picking and assembling parts, tightening nuts and screws accurately, and sorting industrial components directly from moving conveyor belts.

This cross-layer investment distribution across compute, agents and robots reshapes procurement and development calculations in the Arab region, particularly in the Gulf, Egypt and the Levant. For data-centre managers and regional cloud operators who are pouring massive infrastructure investments, the maturation of alternative linking technologies to copper or radio frequencies means the possibility of lowering cooling costs and energy consumption when scaling processor clusters. For banks and large corporations seeking transaction automation, the emergence of specialised document models and open-weight agent models gives local engineering teams the ability to deploy sovereign internal solutions without relying exclusively on closed-cloud APIs. In logistics and light manufacturing, reliance on robots shifts from fixed-arm stages to foundational robotic models capable of handling changing warehouse environments.

The final step rests on a broad base that Nvidia has built in South Korea, where the Nvidia Inception programme includes more than a thousand local start-ups benefiting from preferential pricing, technical expertise and networking opportunities with investors. The direct result of this expansion confirms that the battle for AI sovereignty is not decided merely by acquiring chips, but by controlling integrated supply chains from physical interconnects through agent models to precise mechanical automation.

Don't miss the next story

Subscribe for updates