Nvidia Opens Its Racks to a Korean Chip

SEOUL, Nov. 9 — Rebellions,Rebellions, a South Korean chip startup, will connect its next-generation Rebel R200 neural processing units to Nvidia’s platform through NVLink Fusion, and SK Telecom plans to deploy the systems in its AI data center, the three companies said on Monday. The plan splits the work of generating an AI model’s response between the two kinds of chip: “NVIDIA GPUs for prefill and Rebellions NPUs for decode,” in the release’s words.Prefill and DecodeThe companies call it disaggregated inference: “assigning different stages of AI response generation to the processor architecture best suited to each task,” to improve throughput and power efficiency, the release says.

Deployment is still a forecast. “Rebellions’ R200 systems are expected to be ready for deployment in 2028,” the release says.

The agreement admits another company’s processor into Nvidia’s racks, but on Nvidia’s terms: over its NVLink interconnect, and in its MGX rack design. NVLink Fusion, the release says, lets chipmakers “combine specialized silicon with NVIDIA’s high-bandwidth, low-latency scale-up networking.” “With NVLink Fusion, Korea’s chipmakers can connect their custom ASICs to NVIDIA’s AI factory platform, tightly coupling their processors with NVIDIA compute to run AI workloads,” Jensen Huang, Nvidia’s founder and chief executive, said in the release.NVDANasdaq178.67▼7.97−4.3%Nvidia · Monday’s session

Seoul is backing the work. Korea’s Ministry of Science and ICT is supporting collaboration between Nvidia and Korean NPU companies, “beginning with Rebellions and SK Telecom,” the release says. “This collaboration with NVIDIA marks a significant step forward, as it enables Korean-developed NPUs to enter the computing infrastructure market alongside NVIDIA, a global leader in AI computing, through NVLink Fusion,” Bae Kyung-hoon, the deputy prime minister and minister of science and ICT, said in the release. SK Telecom, one of Rebellions’ backers, plans to connect the systems to its own models, including A.X K2, and to public-facing services under the government’s “AI for All” project.RebellionsRebellions is based in South Korea and operates in the United States. Its flagship chip, the Rebel100, is built for inference. Its backers are Aramco, Arm, Kindred Ventures, KT, Mirae Asset Group, Samsung, SK Hynix and SK Telecom, the release says.

For the other Korean chipmakers, the door is only ajar: the ministry’s wider collaboration with Nvidia “lays the groundwork” for them “to explore participation” in NVLink Fusion, Nvidia wrote on its blog. SK Telecom’s own caution, printed with the release, is that actual results may differ materially because of “technological and operational challenges” and the “availability of necessary resources.”

The release went out at midnight Eastern time, 2 p.m. in Seoul, on the first day of Nvidia AI Day Seoul, a two-day event at COEX that about 2,000 people were expected to attend, Maeil Business Newspaper reported. Nvidia also announced an investment in Point2, a Korean interconnect startup, with more planned in Motif Technologies, RLWRLD and Upstage; Nvidia gave no amounts.

The day went ahead after Nvidia’s board said on Sunday that it was opening an independent investigation “in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort,” as the paper’s lead story reports; the board said it had “no reason to believe” that national security or Nvidia’s secrets were compromised. Madison Huang, Mr. Huang’s daughter and an Nvidia executive, had planned to visit Rebellions and hold a question-and-answer session with Kakao Mobility; Rebellions said the events were cancelled because of “unavoidable internal circumstances at Nvidia,” Reuters reported.