Key Notes
- Nvidia and SK Telecom plan to integrate Rebellions’ R200 chips into Korean sovereign AI infrastructure.
- The proposed system splits inference between Nvidia GPUs and Rebellions NPUs using NVLink Fusion and MGX.
- R200 systems are expected to be ready for deployment in 2028, with no contract value disclosed.
Nvidia (Nasdaq: NVDA) is teaming up with Korean chipmaker Rebellions and SK Telecom (NYSE: SKM) to develop AI systems that combine Nvidia GPUs with Korean-designed processors, extending its infrastructure platform into South Korea’s push for sovereign AI.
Rebellions plans to connect its next-generation Rebel R200 neural processing units to Nvidia’s NVLink Fusion interconnect and MGX rack architecture. SK Telecom intends to deploy the combined systems in its AI data centers, according to Nvidia’s announcement.
The arrangement offers Nvidia a way to remain part of the infrastructure budget even when customers choose specialized chips from another supplier. For Rebellions, it provides access to established networking, rack designs and manufacturing partners rather than requiring the startup to assemble that ecosystem itself.
GPUs and Korean Chips Split the Inference Work
The companies signed a memorandum of understanding at Nvidia AI Day Seoul at COEX on Monday, November 9, SK Telecom said. The project focuses on inference: running trained AI models to respond to users, rather than training those models from scratch.
The proposed design divides that work into two stages. Nvidia GPUs would handle prefill, the compute-intensive task of processing a user’s prompt. Rebellions’ NPUs would handle decode, generating the response one token at a time, a stage that depends heavily on moving data quickly from memory.
Matching different processors to those tasks is intended to increase throughput and improve energy efficiency. SK Telecom plans to test the integrated system’s performance and reliability in its data centers and assess its use in commercial services.
For an operator selling AI services, the economics come down to how many responses a system can deliver within its power and equipment budget. A specialized chip has commercial value only if the savings survive the work of connecting it to the rest of the data center.
Nvidia’s Infrastructure Bet Goes Beyond Its Own Chips
NVLink Fusion is central to that strategy. Nvidia introduced the platform in May 2025 to let other companies connect custom processors to its computing infrastructure. The Korean agreement applies that approach to a domestic chip developer and a local operator.
The pitch extends beyond a fast connection between chips. Nvidia’s platform documentation describes a shared framework for networking, power, cooling, system software and manufacturing supply chains. Common rack designs can support different processors without requiring operators to develop a separate physical infrastructure for each one.
That creates a different competitive proposition from selling every accelerator in a facility. A customer may choose a specialized processor for part of an AI workload while continuing to use Nvidia GPUs, interconnects and system architecture around it.
The Stakes for Nvidia Stock
Data centers dominate Nvidia’s financial results. The company reported $89 billion of Data Center revenue in its second quarter of fiscal 2027, up 117% from a year earlier, out of total revenue of $96.2 billion. Its third-quarter outlook called for $108 billion in sales, plus or minus 2%.
For NVDA shareholders, the Korean collaboration shows how Nvidia is trying to preserve its role as computing systems become more varied. It could broaden demand for the company’s infrastructure even where a local processor takes on some of the inference work.
The announcement disclosed no contract value, chip quantities or revenue commitment. SK Telecom described a development and validation MOU, so there is no announced order size to add to Nvidia’s sales forecast.
SK Telecom Connects Infrastructure and Services
SK Telecom wants to connect the hardware to its own foundation models, including A.X K2, and public-facing services under Korea’s “AI for All” project. The collaboration has support from the Ministry of Science and ICT and targets services suited to Korea’s language, industries and public-sector needs.
The operator has wider infrastructure ambitions. In July, it outlined plans for AI data center capacity of up to 15 gigawatts, with an initial 5 gigawatts opening in stages from 2029. It said funding would combine its own investment, strategic partners, long-term customer contracts and project financing.
The July capacity roadmap and Monday’s chip-development agreement have different timetables. SK Telecom has not specified the scale of an R200 installation. Nvidia expects Rebellions’ R200 systems to be ready for deployment in 2028.
