NVIDIA, Rebellions, SK Telecom unveil NVLink Fusion tie-up for Korean AI inference systems

NVIDIA, the Korean chip designer Rebellions and the telecoms operator SK Telecom announced a collaboration on 8 November 2026 to build AI inference systems for Korea around NVIDIA's NVLink Fusion technology. According to NVIDIA's press release, the work has the support of Korea's Ministry of Science and ICT (MSIT).

What each company is doing

Under the agreement, Rebellions will link its Rebel R200 neural processing units (NPUs) to NVIDIA's platform through NVLink Fusion, and will build rack-scale systems on NVIDIA's MGX rack architecture. NVIDIA describes NVLink Fusion as a way for chipmakers to pair their own specialised silicon with NVIDIA's high-bandwidth, low-latency scale-up network and its wider infrastructure ecosystem. Within each rack, the R200 chips will be linked to one another over NVLink.

NVIDIA says Rebellions is still developing the R200, a next-generation NPU it describes as tuned for low-latency, energy-efficient inference on frontier models, including mixture-of-experts models with multiple trillions of parameters. NVIDIA adds that through the MGX ecosystem, Rebellions can draw on existing suppliers, from original design manufacturers to component makers.

SK Telecom plans to install the resulting systems in its own AI data centre, using the standardised MGX rack design. The operator plans to link them to its own foundation models, which its chief executive, Jung Jai-hun, says include A.X K2, and to its services for the public, among them those built under the government's "AI for All" project.

Shared rack design and split inference

NVIDIA says the MGX architecture lets GPU and NPU systems use common data centre design elements, covering the rack footprint, how power is delivered, and cooling, and that this can support infrastructure planning while Rebellions develops its processor and system.

The three companies also plan to develop what the release calls disaggregated inference, in which different stages of generating an AI response go to whichever processor suits them best. Their planned work has NVIDIA GPUs handling prefill and Rebellions NPUs handling decode, with the stated aim of raising throughput and power efficiency.

Timing and government support

According to the release, the R200 systems from Rebellions are expected to be ready for deployment in 2028. The release quotes Korea's Deputy Prime Minister and science minister, Bae Kyung-hoon, saying the project allows Korean-developed NPUs to reach the market for computing infrastructure through NVLink Fusion, and that the government will support it and deepen cooperation with NVIDIA in areas that include physical AI and GPU supply. NVIDIA says the ministry is backing joint work pairing NVIDIA with Korean NPU makers, starting with the Rebellions and SK Telecom project.

A separate NVIDIA blog post of the same date says NVIDIA and the ministry are widening their cooperation across Korea's AI ecosystem, and that this prepares the ground for Korean makers of NPUs and XPUs to consider taking part in the NVLink Fusion ecosystem. The release describes Rebellions as a builder of accelerators for AI inference, aimed at enterprises and countries, headquartered in South Korea with operations in the United States, and lists its backers as Aramco, Arm, Kindred Ventures, KT, Mirae Asset Group, Samsung, SK Hynix and SK Telecom.