Nvidia, Rebellions and SK Telecom Link Korean AI Chips to Nvidia Racks via NVLink Fusion

The deal shows how Nvidia is trying to make its interconnect technology the meeting point for rival silicon rather than lose ground to it. NVLink Fusion lets other chipmakers attach custom processors to Nvidia's high-bandwidth scale-up networking and its MGX rack designs. For Rebellions, it offers a route into large data centres alongside Nvidia hardware instead of having to compete with it head-on. For Seoul, it ties a homegrown chip company to the government's push for sovereign AI, meaning computing capacity, models and services built and run in Korea.

Under the agreement, Rebellions will use NVLink to link R200 chips within a rack and build its systems on Nvidia's MGX architecture, which also gives it access to Nvidia's existing network of original design manufacturers and component suppliers. According to Nvidia's release, the R200 is designed for low-latency, energy-efficient inference on frontier models, including mixture-of-experts models with several trillion parameters. The release says the R200 systems are expected to be ready for deployment in 2028.

The companies also plan to split inference work between the two types of chip. Nvidia GPUs would handle prefill, the stage in which a model processes the incoming prompt, while Rebellions' NPUs would handle decode, the token-by-token generation of the answer. The partners said this should improve inference throughput and power efficiency. Because GPU and NPU racks would share the MGX footprint, power delivery and cooling designs, SK Telecom could plan its data centre around both.

SK Telecom said it would connect the infrastructure to its own foundation models, including A.X K2, and to public-facing services built under the government's "AI for All" project. "By linking SKT's AI data centers, proprietary AI models including A.X K2, and public-facing AI services developed under the 'AI for All' project, we will strengthen our full-stack sovereign AI competitiveness," said Jung Jai-hun, SK Telecom's president and chief executive.

Sunghyun Park, co-founder and chief executive of Rebellions, called the announcement "a turning point" for sovereign AI and said the company aimed to establish Korean technology "as a core pillar powering AI worldwide."

Nvidia founder and chief executive Jensen Huang is quoted in the release saying that NVLink Fusion lets Korea's chipmakers "connect their custom ASICs to NVIDIA's AI factory platform." The announcement came shortly after Nvidia's board said it was opening an independent investigation into reports that Huang had been the target of a foreign intelligence effort. The board said it had no reason to believe that national security or Nvidia secrets had been compromised.

Bae Kyung-hoon, South Korea's deputy prime minister and minister of science and ICT, said in the release that the collaboration would let Korean-developed NPUs enter the computing infrastructure market alongside Nvidia. He said the government would actively support it and expand cooperation with Nvidia in areas including physical AI and securing GPU supply.

The ministry is backing collaboration between Nvidia and Korean NPU makers "beginning with Rebellions and SK Telecom," according to the release, which suggests other Korean chip designers could join later. No other company was named, and the partners did not disclose financial terms.